diff --git a/sources/harvest-uk-ab2026-2026-10-29/uk_centax/NOTES.md b/sources/harvest-uk-ab2026-2026-10-29/uk_centax/NOTES.md index 912f00ed7..a9ec0184f 100644 --- a/sources/harvest-uk-ab2026-2026-10-29/uk_centax/NOTES.md +++ b/sources/harvest-uk-ab2026-2026-10-29/uk_centax/NOTES.md @@ -1,27 +1,27 @@ # uk_centax: CenTax, Autumn Budget 2026 -This family stages three CenTax publications: +In this family we stage three CenTax publications: - **"Autumn Budget 2026: Initial reaction"** (28 October 2026, centax.org.uk). -- **The two AB2026 briefings it links:** "Taxes at the top: Understanding what high earners pay and options for reform" (23 September 2026), which sets out "reforms available at the 2026 Budget", and "Capital Gains Tax on Entrepreneurs: Should the UK have a 'Unicorn Relief'?" (8 October 2026). +- **The two AB2026 briefings it links.** In "Taxes at the top: Understanding what high earners pay and options for reform" (23 September 2026), CenTax sets out "reforms available at the 2026 Budget". The second is "Capital Gains Tax on Entrepreneurs: Should the UK have a 'Unicorn Relief'?" (8 October 2026). -The source slug is `centax`, as in the AB2025 `uk_centax` family, and `source_model` is `hmrc_administrative_microdata` for CenTax's estimates and descriptive statistics. +We use the source slug `centax`, as in the AB2025 `uk_centax` family, and `source_model` `hmrc_administrative_microdata` for CenTax's estimates and descriptive statistics. -**Working files.** `stage.py`, the file meant wherever these notes say `stage.py`, lives on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/thinktanks-a/uk_centax/stage.py`; this branch carries only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz`. The lane's local primaries folder (`.ab2026-primaries/uk_centax/`) was never committed: the repository carries no primary bytes, so fetch each primary from its `manifest.jsonl` URL and check its sha256. +We keep `stage.py` on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/thinktanks-a/uk_centax/stage.py`. We commit only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz` to this branch, and no primary bytes to the repository. The lane never committed its local primaries folder (`.ab2026-primaries/uk_centax/`), so fetch each primary from its `manifest.jsonl` URL and check its sha256. -`stage.py` holds the rows as data and rewrites `manifest.jsonl` and `claims_staged.jsonl.gz` byte for byte. `stage.py --verify` checks each primary's sha256 and finds every quote in its primary's text (HTML text, or `pdftotext -layout`, whitespace-normalised). Result: 0 problems. +We wrote the rows into `stage.py` as data, and it rewrites `manifest.jsonl` and `claims_staged.jsonl.gz` byte for byte. `stage.py --verify` checks each primary's sha256 and finds every quote in its primary's text (HTML text, or `pdftotext -layout`, whitespace-normalised); it reports 0 problems. -- **Rows:** 112 (own 108, restated 4). -- **Primaries:** 6 (the reaction page in two versions, rule 7). +- Rows: 112 (own 108, restated 4). +- Primaries: 6 (the reaction page in two versions, rule 7). ## Access recipe -Every primary was fetched directly on 2026-10-29 with `curl -sSL` and the browser User-Agent in the brief (HTTP 200). Nothing came from a local copy, and LinkedIn was not used. +The lane fetched every primary directly on 2026-10-29 with `curl -sSL` and the browser User-Agent in the brief (HTTP 200). It took nothing from a local copy and didn't use LinkedIn. -- **The reaction page** was fetched at 16:17Z. It says "This page is being updated live" and carries no published or modified time, so a later fetch may hash differently. -- **The reaction page's 4 November version** (sha256 `5103e06ae3a5831a…`, 271,039 bytes, retrieved 2026-11-04 by review B and matched by the coordinator's own fetch that day) says "This page has finished updating". It adds seven sections the 29 October copy lacks: Big picture; Tax reliefs for founders and employees (with its CGT, BADR, Founders' Relief, EMI2/CSOP and evaluation parts); Company distributions framework. Their markup accounts for the growth from 248,542 bytes. It is a second primary (README rule 7): the 3 earlier rows keep citing the 29 October copy, whose quotes it still prints, and the 2 rows from the new sections cite it. This session did not re-open either version; the two new rows' quotes are the coordinator's, checked against its local copy. -- **The two briefing landing pages** were fetched at 16:30Z. Each prints no figures (0 rows) and links its PDF; they are manifested as the route to the PDFs. -- **The two PDFs** were fetched at 16:31Z and read with `pdftotext -layout`. Locators are printed page numbers; in "Taxes at the top" the Key findings are p. 1. +- The lane fetched the reaction page at 16:17Z. CenTax says "This page is being updated live" and prints no published or modified time, so a later fetch may hash differently. +- Review B retrieved the reaction page's 4 November version on 2026-11-04 (sha256 `5103e06ae3a5831a…`, 271,039 bytes), and the coordinator's own fetch that day matched it. In it CenTax says "This page has finished updating". CenTax added seven sections that the 29 October copy lacks: Big picture; Tax reliefs for founders and employees (with its CGT, BADR, Founders' Relief, EMI2/CSOP and evaluation parts); Company distributions framework. CenTax's markup for those sections accounts for the growth from 248,542 bytes. We treat it as a second primary (README rule 7). On the 3 earlier rows we keep citing the 29 October copy, whose quotes CenTax still prints, and on the 2 rows from the new sections we cite the new version. This session re-opened neither version; the coordinator chose the two new rows' quotes and checked them against its local copy. +- The lane fetched the two briefing landing pages at 16:30Z. CenTax prints no figures on either (0 rows) and links each PDF from it, so we manifest them as the route to the PDFs. +- It fetched the two PDFs at 16:31Z and read them with `pdftotext -layout`. We use printed page numbers as locators; in "Taxes at the top" CenTax puts the Key findings on p. 1. ## Rows by primary @@ -38,18 +38,19 @@ Every primary was fetched directly on 2026-10-29 with `curl -sSL` and the browse | measure_key | rows | |---|--:| -| `null`: 53 with a proposed key, 56 context (descriptive HMRC statistics, current-policy childcare behaviour, OBR CGT forecasts) | 109 | +| the seven option keys below | 53 | +| `null`: context (descriptive HMRC statistics, current-policy childcare behaviour, OBR CGT forecasts) | 56 | | `ab2026__pension_tax_allowance` (CenTax's 53% marginal rate) | 1 | | `ab2026__hmrc_tax_gap_package` (the timely-payment component, restated from the OBR, `row_type: measure_component`, README rule 9) | 1 | | `ab2026__distributions_framework_modernised` (the £800m, restated from Table 6.1 line 49) | 1 | -## Proposed measure keys +## Option keys -The Budget took none of the options below. Founders' Relief (Table 6.1 line 26) and the BADR cut (line 60) are a different CGT design from CenTax's package. Each option is CenTax's own design, in CenTax's words: +On 29 October we gave the 53 option rows `measure_key: null` and proposed the seven keys below, and in tranche 2 we registered them. The government took none of these options in the Budget. It chose Founders' Relief (Table 6.1 line 26) and the BADR cut (line 60), a different CGT design from CenTax's package. CenTax designed each option, and we give each in CenTax's words: -| proposed key | scope | rows | +| key | scope | rows | |---|---|--:| -| `ab2026_option__cgt_reform_package` | "Structural reform of CGT and equalising rates with Income Tax": equalise CGT rates with Income Tax rates; an investment allowance for the risk-free rate; remove uplift at death; rebasing on arrival with deemed disposal on departure. "We use administrative data from HMRC to cost the first four elements"; the fifth, more generous loss relief, isn't costed. No Table 6.1 line | 10 | +| `ab2026_option__cgt_reform_package` | "Structural reform of CGT and equalising rates with Income Tax": equalise CGT rates with Income Tax rates; an investment allowance for the risk-free rate; remove uplift at death; rebasing on arrival with deemed disposal on departure. "We use administrative data from HMRC to cost the first four elements"; CenTax doesn't cost the fifth, more generous loss relief. No Table 6.1 line | 10 | | `ab2026_option__nics_on_partnership_income` | "Applying an equivalent of Employer NICs to partnership profits": a top-up rate ("6.9% on partnership profits taxed at the additional rate, 7.6% … higher rate, and 9.6% … basic rate"), a £10,500 Partnership allowance and a £5,000 annual exempt amount per partner. No Table 6.1 line | 24 | | `ab2026_option__nics_on_investment_income` | "Applying an equivalent of Class 1 NICs to investment income": rental income, savings income outside ISAs and other non-dividend investment income (`conditions.income_type`). No Table 6.1 line | 10 | | `ab2026_option__childcare_100k_threshold_removed` | "Removing the £100,000 childcare cliff-edge": make the free childcare entitlement in England available to parents with incomes over £100,000. No Table 6.1 line | 3 | @@ -57,37 +58,37 @@ The Budget took none of the options below. Founders' Relief (Table 6.1 line 26) | `ab2026_option__childcare_100k_threshold_raised` | "Increase the threshold at which entitlement to free childcare is withdrawn", to £125,000 or £150,000 (`conditions.threshold`) | 2 | | `ab2026_option__childcare_100k_entitlement_taper` | "'Taper' the withdrawal of free childcare", at 8p, 18p or 28p per £1 over £100,000 (`conditions.taper_rate`) | 3 | -Only the options short of full removal (rows 101–106) are "costed on an actual ANI basis" (`conditions.costing_basis`); the full removal's £640m and £520m (rows 1, 25 and 99) carry no costing basis. CenTax's actual-ANI threshold itself ("could be set higher (£111,000) … at the same overall fiscal cost") is that option's design and has no row (README rule 8). +CenTax says the options short of full removal (rows 101–106) are "costed on an actual ANI basis", and we record that in `conditions.costing_basis`; we give the full removal's £640m and £520m (rows 1, 25 and 99) no costing basis. CenTax's actual-ANI threshold itself ("could be set higher (£111,000) … at the same overall fiscal cost") forms that option's design, so we staged no row for it (README rule 8). -The £1.6bn timely-payment figure carries `ab2026__hmrc_tax_gap_package` under README rule 9, not a proposed key. It repeats the OBR's printed average (EFO paras 3.4 and 3.13), so it is `restated` (review B, finding 1). More timely payment for Self Assessment is HM Treasury Table 6.3 line 3, a sub-measure of Table 6.1 line 45. The row has `conditions.row_type: measure_component`, with `conditions.component` naming the part. +Under README rule 9 we key the £1.6bn timely-payment figure to `ab2026__hmrc_tax_gap_package`, not a new key. CenTax repeats the OBR's printed average (EFO paras 3.4 and 3.13), so we mark the row `restated` (review B, finding 1). HM Treasury lists more timely payment for Self Assessment as Table 6.3 line 3, a sub-measure of Table 6.1 line 45. We set `conditions.row_type: measure_component` on the row and name the part in `conditions.component`. ## Conventions -- **Tax years:** CenTax gives tax years as the later year ("2022 is tax year 2021-22", Taxes at the top footnote 2). So "2030" is staged as `fy` 2029-30, "2028" as 2027-28, and the descriptive 2022 data as 2021-22. Every such row carries a `year_note`. This matches #207's reading of the same CenTax childcare figures. -- **Sign:** - - Revenue-raising options are `positive = yield to the Exchequer`. - - The childcare options are `positive = cost to the Exchequer`, except "raise £20 million" (28p taper), which is staged as +2e7 with `positive = yield to the Exchequer`: values are not re-signed (AB2025 README; changed on 3 November, review finding 4). -- **"Central":** CenTax's central scenario (`conditions.scenario: central`). It uses "elasticities drawn from the best available international evidence" for CGT, its "central revenue estimate, after accounting for behavioural responses" for partnership NICs, and its central bunching scenario for childcare. The investment-income NICs figures are static upper bounds ("naturally at the upper bound", footnote 13), so they are `range_high` with `basis: static`. -- **Units:** - - Shares are fractions (`share`). - - Effective and marginal rates are `percent`; rate gaps are `percentage_points`. No top-up rate is staged: the partnership top-up rates are the option's design (below). - - The CGT package's "increase revenue by 62%" is `revenue_change` in `percent` of baseline CGT revenue. - - Income groups follow the AB2025 names (`top_10pct`, `top_1pct`, `top_0.1pct`, `top_0.01pct`, `decile_10`, `bottom_50pct`); percentile bands are `p50_p60` … `p99.5_p99.9`. Single percentiles are `median`, `p99` and `p99.99`, and p. 11's "by around 2.8 percentage points on average" is `top groups, on average`. -- **Attribution:** - - CenTax's model estimates are `own` / `different_model`. - - Descriptive HMRC statistics are `own` / `administrative_fact`, as in AB2025. - - The OBR's £32bn and £16bn CGT forecasts are `restated`. - - The £1.6bn timely-payment average repeats the OBR's printed figure, so it is `restated` with `source_model: obr_efo_forecast` and `conditions.originator` naming EFO paras 3.4 and 3.13 (review B, finding 1). - - The £800m for the distributions framework (4 November version) restates Table 6.1 line 49: `restated`, `hmt_costing_obr_certified`. - - CenTax's 53% marginal rate is `own` with `source_model: arithmetic`. - - The "over half of CGT payers" claim (4 November version) is CenTax's own, about its own CGT package: `own`, `share_with_tax_cut`, value 0.5, `range_low`, medium confidence (no year printed; staged with the package's 2030 costing). -- **Context rows:** the descriptive 2022 tax statistics and the current-policy childcare behaviour (bunching counts, response shares, partners leaving work) have no measure key and no proposal. The option keys carry only the options' costings and the distribution of their revenue or affected taxpayers. - -## Proposed metrics (registered Metrics used where they fit) - -Registered: `revenue_change`, `revenue_share` (also for "paid 60% of Income Tax and NICs"), `income_share`, `income_statistic`, `average_tax_rate`, `affected_share`, `share_with_tax_cut`, `participation_rate`, `tax_collections`. - -Proposed, with CenTax's own words: +- **Tax years.** CenTax names each tax year by its later year ("2022 is tax year 2021-22", Taxes at the top footnote 2). So we stage "2030" as `fy` 2029-30, "2028" as 2027-28 and the descriptive 2022 data as 2021-22, and give every such row a `year_note`. We read the same CenTax childcare figures this way in #207. +- **Sign.** + - We stage revenue-raising options as `positive = yield to the Exchequer`. + - We stage the childcare options as `positive = cost to the Exchequer`. We stage the one exception, the 28p taper's "raise £20 million", as +2e7 with `positive = yield to the Exchequer`, because we don't re-sign values (AB2025 README; changed on 3 November, review finding 4). +- **"Central".** We record CenTax's central scenario as `conditions.scenario: central`. For CGT CenTax uses "elasticities drawn from the best available international evidence", for partnership NICs its "central revenue estimate, after accounting for behavioural responses", and for childcare its central bunching scenario. CenTax gives the investment-income NICs figures as static upper bounds ("naturally at the upper bound", footnote 13), so we stage them as `range_high` with `basis: static`. +- **Units.** + - We give shares as fractions (`share`). + - We give effective and marginal rates in `percent` and rate gaps in `percentage_points`. We stage no top-up rate, because the partnership top-up rates form the option's design (below). + - We stage the CGT package's "increase revenue by 62%" as `revenue_change` in `percent` of baseline CGT revenue. + - We name income groups as in AB2025 (`top_10pct`, `top_1pct`, `top_0.1pct`, `top_0.01pct`, `decile_10`, `bottom_50pct`), and percentile bands `p50_p60` … `p99.5_p99.9`. We name single percentiles `median`, `p99` and `p99.99`, and p. 11's "by around 2.8 percentage points on average" `top groups, on average`. +- **Attribution.** + - We mark CenTax's model estimates `own` / `different_model`. + - We mark descriptive HMRC statistics `own` / `administrative_fact`, as in AB2025. + - We mark the OBR's £32bn and £16bn CGT forecasts `restated`. + - CenTax repeats the OBR's printed figure for the £1.6bn timely-payment average, so we mark it `restated`, with `source_model: obr_efo_forecast`, and name EFO paras 3.4 and 3.13 in `conditions.originator` (review B, finding 1). + - CenTax restates Table 6.1 line 49 for the £800m for the distributions framework (4 November version), so we mark it `restated`, `hmt_costing_obr_certified`. + - We mark CenTax's 53% marginal rate `own`, with `source_model: arithmetic`. + - CenTax makes the "over half of CGT payers" claim (4 November version) on its own account, about its own CGT package. We mark it `own`, `share_with_tax_cut`, value 0.5, `range_low`, medium confidence; CenTax prints no year, so we stage it with the package's 2030 costing. +- **Context rows.** We gave no measure key and no proposal to the descriptive 2022 tax statistics or the current-policy childcare behaviour (bunching counts, response shares, partners leaving work). We key to the options only their costings and the distribution of their revenue or affected taxpayers. + +## Proposed metrics (registered metrics where they fit) + +We use these registered metrics: `revenue_change`, `revenue_share` (also for "paid 60% of Income Tax and NICs"), `income_share`, `income_statistic`, `average_tax_rate`, `affected_share`, `share_with_tax_cut`, `participation_rate`, `tax_collections`. + +We propose these, with CenTax's own words: - **`marginal_tax_rate` (percent):** "a new marginal rate of 53%". - **`effective_tax_rate_gap` (percentage points):** "the gap between headline rates and the average EATRs". @@ -95,59 +96,64 @@ Proposed, with CenTax's own words: - **`behavioural_response_share` (share):** "fewer than one in ten employees with eligible children responded to the cliff-edge". - **`workforce_exit_share` and `female_share` (share):** "3% of lower paid partners … leaving the workforce, with 90% of those partners being women". - **`partner_not_in_work_share` (share):** "the share of parents whose partner is not in work is about 6%". -- **`parents_bunching` (count):** "from 1,100 in 2022 to 11,980 by 2030" (rows 26–27; the label of the repository's childcare adapter, which drops these counts as "a behavioural-model count of parents" with no registered metric; review B, finding 3). +- **`parents_bunching` (count):** "from 1,100 in 2022 to 11,980 by 2030" (rows 26–27; we gave these counts this name in #207's `data/externals/centax-childcare-100k.json`. The repository's childcare ingester drops them as "a behavioural-model count of parents" with no registered metric; review B, finding 3). ## Not staged, and why -- **"Around four times as many parents would be bunching"** (p. 25), the halved-unresponsiveness sensitivity: a scenario ratio, not an estimate of an option's effect (added on 3 November, review finding 6). - -- **Duplicates within one primary,** staged once: - - in "Taxes at the top": 62%, 88%, £19.7bn, £2.1bn, 98%, 1,100, 90% and £640m (Key findings, repeated on pp. 17-27); 37%, 34% and 30% (p. 9); 17%→12%, 33%→36% and 37%→33% (p. 13); 40.5%, 60%, 43.9%, 61.5%, 47%, 20% and 23% (Conclusion); - - in the Unicorn Relief PDF: "almost £20 billion" restates the £19.7bn in the same footnote. -- **Design, not estimates** (README rule 8): - - the reaction page's "marginal tax rate of 33.3%" (the Pension Tax Allowance's £1-for-£3 withdrawal rate, restated as a percentage; kept in the 53% row's conditions) and "a 100% marginal tax rate" (the Carer's Allowance taper of £1 for every £1, restated); +- **"Around four times as many parents would be bunching"** (p. 25): CenTax gives this for its halved-unresponsiveness sensitivity, as a scenario ratio, not an estimate of an option's effect. We added this entry on 3 November, after review finding 6. + +- We staged each duplicate within one primary once: + - in "Taxes at the top": + - Key findings, repeated on pp. 17-27: 62%, 88%, £19.7bn, £2.1bn, 98%, 1,100, 90% and £640m; + - p. 9: 37%, 34% and 30%; + - p. 13: 17%→12%, 33%→36% and 37%→33%; + - Conclusion: 40.5%, 60%, 43.9%, 61.5%, 47%, 20% and 23%; + - in the Unicorn Relief PDF: CenTax's "almost £20 billion" restates the £19.7bn in the same footnote. +- Under README rule 8 we left out these figures, which set an option's or a measure's design and estimate nothing: + - the reaction page's "marginal tax rate of 33.3%", the Pension Tax Allowance's £1-for-£3 withdrawal rate restated as a percentage, which we keep in the 53% row's conditions; + - its "a 100% marginal tax rate" (the Carer's Allowance taper of £1 for every £1, restated); - Taxes at the top's partnership top-up rates of 6.9%, 7.6% and 9.6% (the option's rates, quoted in its scope above) and its £111,000 actual-ANI threshold. -- **Policy parameters** (also design under rule 8): +- We left out policy parameters, which we also treat as design under rule 8: - in the reaction page: the £12,570 Personal Allowance; the 24%, 44% and 49% rates; the £12,000 cash ISA cap from AB2025; the Direct Recovery of Debts thresholds (£1,000, £5,000, £10,000) and dates; - - in "Taxes at the top": Employer NICs at 15%, its "effective rate of Employer NICs is 13.04%", the £10,500 and £5,000 allowances, footnote 11's rates, and Appendix A's rates. -- **The reaction page's worked mechanics:** the payments-on-account example (a £10,000 liability, two £5,000 payments) explains the existing regime; it is not a figure about the Budget. -- **Charts without printed values:** Taxes at the top Figures 1-7, B1 and B2. -- **Unicorn Relief's illustrations and design parameters:** + - in "Taxes at the top": Employer NICs at 15%, CenTax's "effective rate of Employer NICs is 13.04%", the £10,500 and £5,000 allowances, footnote 11's rates, and Appendix A's rates. +- We left out the reaction page's worked mechanics. CenTax uses the payments-on-account example (a £10,000 liability, two £5,000 payments) to explain the existing regime, not to give a figure about the Budget. +- We left out charts on which CenTax prints no values: Taxes at the top Figures 1-7, B1 and B2. +- We left out Unicorn Relief's illustrations and design parameters: - the risk-aversion example (£100,000 v £1 million at 10%); - footnote 27's net-of-tax returns on a £100m gain (76% → 76.06% → 77.4%); - the £20m break-even example at a 15% rate; - the taper example (45% → 44% per £1m above £750m); - the £750m / US$1bn threshold; - EMI and SEIS limits. - These illustrate design choices rather than estimate outcomes, the same test AB2025 uk_centax applied ("thresholds/equivalences, not scoreable values"). The report says it does not cost Unicorn Relief ("It is difficult to model how much Unicorn Relief would cost"). -- **The September 2026 childcare report** ("Removing the £100,000 childcare cliff-edge: Impacts and cost of reform", linked from the reaction page) is not re-staged. Its Table 4.2 is already in `data/externals/centax-childcare-100k.json` (#207). That file differs from this family in three ways (thinktanks-a review, finding 3): - - **sign:** #207 stores the full removal's central net fiscal cost as −640m, negative for a cost; this family stores +640m with `positive = cost to the Exchequer`; - - **coverage:** #207 holds only Table 4.2's full removal; the cheaper options (rows 101–106) come only from "Taxes at the top"; - - **bunching counts:** row 27 has "11,980 by 2030", where #207's 2030 central "Number Bunching, Before Removal" is 16,280. This may be a different concept, or CenTax may differ between its two publications; row 27's note says so. The £520m and £640m staged here are CenTax repeating its own central net fiscal cost in AB2026 publications; the ingest should treat them as the same estimate as #207's. -- **LinkedIn:** CenTax's 28 October LinkedIn post (seed inventory) needs a login; not used. -- **The 4 November version's design figures** (README rule 8): the BADR lifetime limit's cut from £1 million to £500,000, Founders' Relief's 10% rate on up to £10 million of gains, and the EMI, EMI2 and CSOP limits. They are the measures' parameters, not estimates. Its "Big picture" section repeats the Budget's headline figures and holds no CenTax estimate. -- **"Taxes at the top", pp. 2–3:** "around two-thirds of total tax revenue" (a description of the tax base, not an estimate about the Budget or an option) and the "over 100%" marginal rates at the childcare cliff edge (the current regime's design, restated). + CenTax uses these to illustrate design choices, not to estimate outcomes, the same test we applied in AB2025 uk_centax ("thresholds/equivalences, not scoreable values"). CenTax says it doesn't cost Unicorn Relief ("It is difficult to model how much Unicorn Relief would cost"). +- We didn't re-stage the September 2026 childcare report ("Removing the £100,000 childcare cliff-edge: Impacts and cost of reform", linked from the reaction page). We already store its Table 4.2 in `data/externals/centax-childcare-100k.json` (#207). That file differs from this family in three ways (thinktanks-a review, finding 3): + - **sign:** in #207 we store the full removal's central net fiscal cost as −640m, negative for a cost; in this family we store +640m with `positive = cost to the Exchequer`; + - **coverage:** #207 covers only Table 4.2's full removal; we take the cheaper options (rows 101–106) only from "Taxes at the top"; + - **bunching counts:** in row 27 we stage "11,980 by 2030", where CenTax gives 16,280 as #207's 2030 central "Number Bunching, Before Removal". CenTax may mean a different concept, or may differ between its two publications, and we say so in row 27's note. In the £520m and £640m we stage here, CenTax repeats its own central net fiscal cost in AB2026 publications; at ingest we score them as CenTax's claims on `ab2026_option__childcare_100k_threshold_removed` (rows 1, 25 and 99). We don't link them to #207's estimate, so that link stays open. +- **LinkedIn:** CenTax's 28 October LinkedIn post (seed inventory) needs a login, and we didn't use it. +- We left out the 4 November version's design figures (README rule 8): the BADR lifetime limit's cut from £1 million to £500,000, Founders' Relief's 10% rate on up to £10 million of gains, and the EMI, EMI2 and CSOP limits. CenTax gives them as the measures' parameters, not estimates. In its "Big picture" section CenTax repeats the Budget's headline figures and gives no estimate of its own. +- We left out two passages from "Taxes at the top", pp. 2–3. CenTax uses "around two-thirds of total tax revenue" to describe the tax base, not to estimate anything about the Budget or an option. It restates the "over 100%" marginal rates at the childcare cliff edge as the current regime's design. ## Review fixes, 3 November 2026 -The thinktanks-a review (`harvest-staging/ab2026/reviews/verify-thinktanks-a.md` on the staging branch) approved this family with fixes. Each sits in an explicit block at the end of `stage.py`'s rows. No primary was re-opened (the session's network policy refused centax.org.uk); no printed value changed. +In its review (`harvest-staging/ab2026/reviews/verify-thinktanks-a.md` on the staging branch), the thinktanks-a reviewer approved this family with fixes. We put each in an explicit block at the end of `stage.py`'s rows. We re-opened no primary, because the session's network policy refused centax.org.uk, and changed no printed value. -- **Finding 1:** rows 84–86 (£3bn, £0.6bn and £0.4bn, p. 22) are the parts of row 24's "up to £4 billion" (3 + 0.6 + 0.4 = 4.0). They carry `conditions.row_type: measure_component` and `conditions.component` naming the income type, so an ingest never adds them to the total. -- **Finding 2:** confirmed. Row 2's derivation cites Table 6.3 row 3 at 1,355, 1,605 and 1,815 (£m, 2029-30 to 2031-32); `uk_hmt_redbook` stages exactly those cells. -- **Finding 3:** the #207 file's three differences are recorded under "Not staged" and in row 27's note. -- **Finding 4 (optional, applied):** row 105 is +2e7 with `positive = yield to the Exchequer` and normalisation "£ million × 1e6". -- **Finding 5 (optional, applied):** row 18 (23%, a 25th-percentile EATR) uses the registered `average_tax_rate`, as rows 7–15 do. Rows 19–20 put `revenue_change` in `percent`; tranche 2 must not read them as £. -- **Finding 6:** p. 25's "around four times as many parents would be bunching" in the halved-unresponsiveness scenario is a scenario ratio and is listed under "Not staged". -- **Finding 7:** rows 20, 58, 108 and 109 are CenTax's 2024-era estimates at fy 2025-26 under the CGT option key; their `scenario` condition says so, for tranche 2. -- **Verification:** the contract checker reports 110 rows, 0 problems, and a re-run of `stage.py` reproduces both files byte for byte. COUNTS.json pins the gzip's sha256. +- Finding 1: CenTax prints £3bn, £0.6bn and £0.4bn (p. 22; rows 84–86) as the parts of its "up to £4 billion" (row 24; 3 + 0.6 + 0.4 = 4.0). We set `conditions.row_type: measure_component` on them and name the income type in `conditions.component`, so an ingest never adds them to the total. +- Finding 2: the reviewer confirmed it. Row 2's derivation cites Table 6.3 row 3 at 1,355, 1,605 and 1,815 (£m, 2029-30 to 2031-32), and we stage exactly those cells in `uk_hmt_redbook`. +- Finding 3: we record the #207 file's three differences under "Not staged" and in row 27's note. +- Finding 4 (optional, applied): we stage row 105 as +2e7 with `positive = yield to the Exchequer` and normalisation "£ million × 1e6". +- Finding 5 (optional, applied): for row 18 (23%, a 25th-percentile EATR) we use the registered `average_tax_rate`, as on rows 7–15. On rows 19–20 we put `revenue_change` in `percent`. At ingest we drop row 19 (62%) as a ratio and row 20 (88%) as CenTax's earlier estimate restated, so neither becomes a claim. +- Finding 6: CenTax gives p. 25's "around four times as many parents would be bunching" in the halved-unresponsiveness scenario as a scenario ratio, and we list it under "Not staged". +- Finding 7: we stage CenTax's 2024-era estimates at fy 2025-26 under the CGT option key in rows 20, 58, 108 and 109, and say so in their `scenario` condition. At ingest we drop row 20 and score rows 58, 108 and 109 on `ab2026_option__cgt_reform_package`, with the `scenario` condition on each claim. +- The contract checker reports 110 rows, 0 problems, and when we re-run `stage.py` it reproduces both files byte for byte. COUNTS.json pins the gzip's sha256. ## Review fixes, 4 November 2026 -Review B (PR #231, issue comment 6635498758) approved this family with fixes, and the coordinator's brief settled finding 2. They sit in an explicit block after the 3 November fixes in `stage.py`. This session re-opened no primary. +Review B (PR #231, issue comment 6635498758) approved this family with fixes, and the coordinator settled finding 2 in the brief. We put the fixes in an explicit block after the 3 November fixes in `stage.py`. This session re-opened no primary. -- **Finding 1:** row 2 (£1.6bn) is `restated`, `source_model: obr_efo_forecast`, with `conditions.originator` "OBR, Economic and fiscal outlook October 2026, paras 3.4 and 3.13" in place of `derived_from`. -- **Finding 2 (coordinator's call):** the reaction page's 4 November version is a second primary (`5103e06a…`). Two rows are staged from it (rows 110–111): the £800m in 2029-30 for the distributions framework, `restated` on `ab2026__distributions_framework_modernised`; and "over half of CGT payers" would pay less tax under CenTax's package, `own` on `ab2026_option__cgt_reform_package`, 0.5 `range_low`. The new sections' design figures are listed under "Not staged". -- **Finding 3:** rows 26–27 (the bunching counts) drop `affected_count` for `proposed_metric: parents_bunching`. -- **Finding 4:** rows 87–92 are `measure_component`s of the investment-income option, `component` = the row's `income_type`. -- **Finding 5:** Table 1's share-of-affected-population rows (67, 69, …, 83) carry `conditions.denominator: "taxpayers affected by the reform"`. -- **Finding 6:** the costing-basis sentence names rows 101–106 only; no top-up rate is staged; the income-group list adds `median`, `p99`, `p99.99` and `top groups, on average`; row 0 has a `year_note`; pp. 2–3's two-thirds and "over 100%" rates are listed under "Not staged". +- Finding 1: we mark row 2 (£1.6bn) `restated`, `source_model: obr_efo_forecast`, with `conditions.originator` "OBR, Economic and fiscal outlook October 2026, paras 3.4 and 3.13" in place of `derived_from`. +- Finding 2, the coordinator's call: we treat the reaction page's 4 November version as a second primary (`5103e06a…`) and stage two rows from it (rows 110–111). We stage the £800m in 2029-30 for the distributions framework as `restated` on `ab2026__distributions_framework_modernised`. We stage CenTax's claim that "over half of CGT payers" would pay less tax under its package as `own` on `ab2026_option__cgt_reform_package`, 0.5 `range_low`. We list the new sections' design figures under "Not staged". +- Finding 3: on rows 26–27 (the bunching counts) we dropped `affected_count` for `proposed_metric: parents_bunching`. +- Finding 4: we mark rows 87–92 as `measure_component`s of the investment-income option, with `component` = the row's `income_type`. +- Finding 5: on Table 1's share-of-affected-population rows (67, 69, …, 83) we set `conditions.denominator: "taxpayers affected by the reform"`. +- Finding 6: we name only rows 101–106 in the costing-basis sentence and say we stage no top-up rate. We added `median`, `p99`, `p99.99` and `top groups, on average` to the income-group list, gave row 0 a `year_note`, and list pp. 2–3's two-thirds and "over 100%" rates under "Not staged". diff --git a/sources/harvest-uk-ab2026-2026-10-29/uk_devolved_govts/NOTES.md b/sources/harvest-uk-ab2026-2026-10-29/uk_devolved_govts/NOTES.md index 00288945e..8b1d8a042 100644 --- a/sources/harvest-uk-ab2026-2026-10-29/uk_devolved_govts/NOTES.md +++ b/sources/harvest-uk-ab2026-2026-10-29/uk_devolved_govts/NOTES.md @@ -1,26 +1,26 @@ # uk_devolved_govts — devolved governments, Autumn Budget 2026 -**Working files.** `stage.py`, which writes `manifest.jsonl` and `claims_staged.jsonl.gz`, lives on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/thinktanks-b/uk_devolved_govts/stage.py`; this branch carries only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz`. The lane's local primaries folder (`.ab2026-primaries/`) was never committed: the repository carries no primary bytes, so fetch each primary from its `manifest.jsonl` URL and check its sha256. +We keep `stage.py`, which writes `manifest.jsonl` and `claims_staged.jsonl.gz`, on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/thinktanks-b/uk_devolved_govts/stage.py`. We commit only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz` to this branch, and no primary bytes to the repository. The lane never committed its local primaries folder (`.ab2026-primaries/`), so fetch each primary from its `manifest.jsonl` URL and check its sha256. -**19 rows from two primaries.** Rows 0–13 (0-based) are the Welsh Government's (source `welsh_government`); rows 14–18 are the Scottish Government's (source `scottish_government`). The Scottish rows were staged in `uk_fai_sfc` on 29 October. On 3 November the thinktanks-b review's finding 7 moved them here, so that both devolved governments sit in one family; `uk_fai_sfc` is left with no rows and is not carried (see the harvest README). Conventions follow the AB2025 `uk_fai_sfc` block-grant rows (`source_model: arithmetic`). +We stage **19 rows from two primaries.** Rows 0–13 (0-based) come from the Welsh Government (source `welsh_government`), and rows 14–18 from the Scottish Government (source `scottish_government`). The lane staged the Scottish rows in `uk_fai_sfc` on 29 October. On 3 November, after finding 7 of the thinktanks-b review, we moved them here so that both devolved governments share one family; we left `uk_fai_sfc` with no rows and don't commit it (see the harvest README). We follow the conventions of the AB2025 `uk_fai_sfc` block-grant rows (`source_model: arithmetic`). ## Changes on 3 November 2026 (thinktanks-b review) -- **Finding 2:** the Welsh Table 1 Resource and Capital cells (rows 1–7) carry `conditions.row_type: "measure_component"` with their `component`; the four Total cells and the £448m headline (rows 0 and 8–11) carried `row_type: "total"` and no `component` (superseded on 4 November, above: all twelve are now parts of the three-government package). An ingest can no longer add parts to their total. -- **Finding 3:** the Welsh package key names its scope, not the nation: `ab2026__package_barnett_consequentials`, shared with HM Treasury's figures for the same scope in `uk_hmt_redbook`. Nation is `conditions.geography`. -- **Finding 4:** the seven real-terms settlement rows (Welsh rows 12–13, Scottish rows 14–18) use `proposed_metric: devolved_funding_real_terms_change`. AB2025's `block_grant_effect` means a policy's effect on the grant, positive for an increase; these rows are a year-on-year real-terms change in the whole settlement, positive for a reduction. -- **Finding 7:** the Scottish Government's rows moved here (above). Both slugs are now in the harvest README's slug table. -- **`component` on context rows:** the seven real-terms rows named their budget (Resource, Capital and so on) in `component`, which rule 9 reserves for parts of a measure. They now use `budget`, by the reasoning of the review's finding 6. -- **Attribution (coordinator check under finding 3):** the Welsh rows stay `own`. The page cites no source, and the coordinator's settled calls do not change this; HM Treasury's own consequentials are staged as HM Treasury's in `uk_hmt_redbook`. +- Finding 2: we gave the Welsh Table 1 Resource and Capital cells (rows 1–7) `conditions.row_type: "measure_component"` with their `component`. We gave the four Total cells and the £448m headline (rows 0 and 8–11) `row_type: "total"` and no `component`. On 4 November we superseded that, and all twelve now form parts of the three-government package (see "Changes on 4 November 2026", finding 1). An ingest can no longer add parts to their total. +- Finding 3: we named the Welsh package key for its scope, not the nation: `ab2026__package_barnett_consequentials`, which we share with HM Treasury's figures for the same scope in `uk_hmt_redbook`. We record the nation in `conditions.geography`. +- Finding 4: we gave the seven real-terms settlement rows (Welsh rows 12–13, Scottish rows 14–18) `proposed_metric: devolved_funding_real_terms_change`. In AB2025 we used `block_grant_effect` for a policy's effect on the grant, positive for an increase. In these rows the two governments instead give a year-on-year real-terms change in the whole settlement, which we store as positive for a reduction. +- Finding 7: we moved the Scottish Government's rows here (above), and we now list both slugs in the harvest README's slug table. +- `component` on context rows: on the seven real-terms rows we had named their budget (Resource, Capital and so on) in `component`, which we use only for parts of a measure (rule 9). Following the reasoning of the review's finding 6, we now name it in `budget`. +- Attribution, which the coordinator checked under finding 3: we keep the Welsh rows `own`. The Welsh Government cites no source, and the coordinator's settled calls do not change this; we stage HM Treasury's own consequentials as HM Treasury's in `uk_hmt_redbook`. -No primary was re-opened on 3 November: the session's network policy refused gov.wales and gov.scot. Every change relabels fields; no printed value changed. +We re-opened no primary on 3 November, because the session's network policy refused gov.wales and gov.scot. We only relabelled fields and changed no printed value. ## Changes on 4 November 2026 (review B) -Review B (PR #231, issue comment 6635498758) approved this family with fixes; the coordinator's brief (section 3.3) settled the Barnett key with review A's K-7(b). The fixes sit in `review_fixes()` in `stage.py`. No primary was re-opened, and no value changed. +Review B (PR #231, issue comment 6635498758) approved this family with fixes, and the coordinator settled the Barnett key with review A's K-7(b) in the brief (section 3.3). We put the fixes in `review_fixes()` in `stage.py`, re-opened no primary and changed no value. -- **Finding 1 (Barnett):** the shared key takes the red book's scope. HM Treasury's £1.3bn (red book row 252) is the total, and each government is a part (rule 9: "a part can be ... a nation"). So rows 0 and 8–11 are `measure_component` with `component: "Welsh Government"`; rows 1–7 name `"Welsh Government: Resource"` or `"Welsh Government: Capital"`; row 0 has `fy: 2029-30`. The proposed-key scope below is the red book's. -- **Finding 2:** row 13's locator says "third sentence", and its note no longer repeats row 12's "next financial year" sentence. Rows 14–18's `source_column` is null, as on rows 12–13: the labels were ours, not printed. The Scottish manifest note no longer names the old `uk_fai_sfc` folder. The sentence on the Welsh package mapping being "unresolved" is corrected. +- Finding 1 (Barnett): we give the shared key the red book's scope. HM Treasury prints the £1.3bn (red book row 252), and we take it as the total and each government as a part (rule 9: "a part can be ... a nation"). So we mark rows 0 and 8–11 `measure_component` with `component: "Welsh Government"`, name rows 1–7 `"Welsh Government: Resource"` or `"Welsh Government: Capital"`, and give row 0 `fy: 2029-30`. In the section on the package key below, we give it the red book's scope. +- Finding 2: we changed row 13's locator to "third sentence", and we no longer repeat row 12's "next financial year" sentence in its note. We set rows 14–18's `source_column` to null, as on rows 12–13, because we wrote those labels and the producer didn't print them. We removed the old `uk_fai_sfc` folder's name from the Scottish manifest note. We corrected the sentence that called the Welsh package mapping "unresolved". ## Coverage @@ -33,114 +33,121 @@ Review B (PR #231, issue comment 6635498758) approved this family with fixes; th | | Paragraphs after Table 1 | 2 | 0.5% resource and almost 10% capital real-terms reductions (context) | | Scottish Government news release | Opening paragraph and paragraph 2 | 5 | £650 million; 1.3% and £550 million (resource); 1.5% and £100 million (capital), all 2027-28 real-terms reductions (context) | -By `measure_key`: **null 19**. Twelve rows carry the proposed key `ab2026__package_barnett_consequentials`, all `measure_component` (Wales is a part of the three-government package); the seven real-terms rows are context with no proposed key. By metric: `block_grant_effect` 12, `devolved_funding_real_terms_change` 7. By unit: `gbp` 15, `percent` 4. Attribution: `own` 19; benchmark `different_model` 19. +By `measure_key`: **`ab2026__package_barnett_consequentials` 12, null 7**. We key the twelve Welsh package rows to it, all `measure_component`, since Wales forms a part of the three-government package. On 3 November we gave these rows the key as a proposal, and in tranche 2 we registered it. We stage the seven real-terms rows as context with no key. By metric we give 12 rows `block_grant_effect` and 7 `devolved_funding_real_terms_change`; by unit, 15 `gbp` and 4 `percent`; we mark all 19 `own`, benchmark `different_model`. -## Proposed measure key +## Package key -`ab2026__package_barnett_consequentials` (12 rows here): the scope `uk_hmt_redbook` gives it, the Barnett consequentials of "decisions confirmed at Budget" for the Scottish, Welsh and Northern Ireland governments together, 2026-27 to 2029-30, excluding the technical reduction from lower employer pension contributions (red book paragraph 4.150). It crosses Table 6.1's England spending lines and has no line of its own. HM Treasury's £1.3bn three-government figure (`uk_hmt_redbook` row 252) is the key's total, and each government's figure is a component. So every Welsh row here is `row_type: measure_component`: the £448m ("an additional £448m of funding over the current spending review period", `fy` 2029-30, cumulative) and the four Table 1 Totals carry `component: "Welsh Government"`, and the Resource and Capital cells `"Welsh Government: Resource"` and `"Welsh Government: Capital"`. The Welsh rows exclude employer-pension-contribution savings, as HM Treasury's do. Compare the Welsh £448m with HM Treasury's £450m for Wales (red book row 254), component with component; never add either to the £1.3bn. +We give `ab2026__package_barnett_consequentials` (12 rows here) the scope we give it in `uk_hmt_redbook`. It covers the Barnett consequentials of "decisions confirmed at Budget" for the Scottish, Welsh and Northern Ireland governments together, 2026-27 to 2029-30. It excludes the technical reduction from lower employer pension contributions (red book paragraph 4.150). We scope the key across Table 6.1's England spending lines, and HM Treasury prints no line of its own for it. HM Treasury prints the £1.3bn three-government figure (`uk_hmt_redbook` row 252), and we take it as the key's total and each government's figure as a component. So we mark every Welsh row here `row_type: measure_component`. We give the £448m ("an additional £448m of funding over the current spending review period", `fy` 2029-30, cumulative) and the four Table 1 Totals `component: "Welsh Government"`, and the Resource and Capital cells `"Welsh Government: Resource"` and `"Welsh Government: Capital"`. The Welsh Government, like HM Treasury, leaves employer-pension-contribution savings out of its figures. Compare the Welsh £448m with HM Treasury's £450m for Wales (red book row 254), component with component; never add either to the £1.3bn. ## Welsh Government ### Access -The [written statement](https://www.gov.wales/written-statement-welsh-government-response-uk-autumn-budget-2026) -was first published and last updated on 28 October 2026. It was fetched fresh -with `curl -sSL` and the exact requested User-Agent: +The Welsh Government first published and last updated the +[written statement](https://www.gov.wales/written-statement-welsh-government-response-uk-autumn-budget-2026) +on 28 October 2026. The lane fetched it fresh with `curl -sSL` and the exact +requested User-Agent: ```text Mozilla/5.0 (Macintosh; Intel Mac OS X 14_0) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/17.0 Safari/605.1.15 ``` -HTTP 200; response Date 2026-10-29T15:22:18Z. The lane saved the unaltered -HTML and response headers in its local primaries folder. BeautifulSoup read the page's paragraphs and actual HTML -table cells, not chart geometry. `shasum -a 256` gives: +The server answered HTTP 200, with response Date 2026-10-29T15:22:18Z. The +lane saved the unaltered HTML and response headers in its local primaries +folder. It read the page's paragraphs and actual HTML table cells with +BeautifulSoup, not chart geometry. `shasum -a 256` gives: ```text 87490552c36d8ffd5787db18127704e467d93599e5708627adb6da7dcfe17c7c ``` -The fetched primary is 26,908 bytes. No local copy was provided, no fallback -was used, and there is no earlier assigned version against which to establish -a change. No requested Welsh page was blocked or unreached. +The fetched primary runs to 26,908 bytes. Nobody provided a local copy, we +used no fallback, and we have no earlier assigned version against which to +establish a change. We reached every requested Welsh page. ### Attribution -These are the Welsh Government's own published settlement accounting and real-terms comparisons. The statement does not credit another producer for these calculations, so no HM Treasury attribution is inferred. They are prospective funding changes rather than administrative outturns or counts. The source does not name a formal model; `arithmetic` describes the settlement accounting. +The Welsh Government publishes these figures as its own settlement accounting and real-terms comparisons. In the statement the Welsh Government credits no other producer for these calculations, so we infer no HM Treasury attribution. The Welsh Government reports prospective funding changes, not administrative outturns or counts. It names no formal model, so we use `arithmetic` for the settlement accounting. ### Welsh numeric and period conventions -- GBP values are the printed £m cells or headline multiplied by one million. - Signed table values are retained: positive means an increase in Welsh - funding. No totals or differences are derived. -- Table columns are fiscal years; `period` is the end year and `fy` retains - the printed year. The £448m headline is cumulative over the four displayed - years, `period: 2030` and `fy: 2029-30` (the horizon's final year), with - that horizon stated explicitly; it is not an annual 2029-30 amount. -- Resource real-terms reduction “next financial year” is 2027-28 given the - publication date. Capital reduction is explicitly by 2029-30. Percentage - values retain the printed magnitude; conditions say positive means a - reduction. “Almost 10%” is preserved in `value_raw`, the upper endpoint - 10 is marked `range_high`, and the qualifier is retained. -- The monetary table has no real-terms basis stated; no deflation is - applied. The percentage context rows retain real terms with no invented - price year. All rows concern Wales. -- `value_kind: point` means a single published figure, not a separately - identified central-model scenario. No row uses `central`. -- Every quote is an exact page fragment or table cell, contains `value_raw` - and is fewer than 15 words. Tables are read by row and then column. - -### Welsh rows not staged - -- Resource 2029-30 is `N/A`, not a number and not zero. -- OBR GDP growth of 1.2% in 2026 and 1.0% in 2027 is restated macro context, - outside this assignment's devolved-consequentials scope; no own Welsh - estimate of a Budget measure is asserted. -- The £100 council-tax-band A-D payment is a Welsh implementation - announcement and policy parameter, with no implementation date, reach, - aggregate cost or estimated household gain printed. It is not attached - to the England-only Treasury line 1 key and is not fabricated into an - impact row. -- The £14 billion rail commitment and £302m already allocated over the - spending review are existing commitments, not newly estimated effects of - AB2026; the statement supplies no isolation of a Budget measure's effect. -- Dates, council tax band letters, and rounding caveats are not estimates. - There are no charts needing bar readings. No inaccessible Welsh page or - unavailable local copy is being used as evidence. +- We multiply the printed £m cells or headline by one million to get GBP + values. We keep signed table values, with positive for an increase in + Welsh funding, and derive no totals or differences. +- The Welsh Government heads the table columns with fiscal years, so we set + `period` to the end year and keep the printed year in `fy`. The Welsh + Government gives the £448m headline as cumulative over the four displayed + years. We record it at `period: 2030` and `fy: 2029-30` (the horizon's + final year) and state that horizon; the £448m doesn't measure + 2029-30 alone. +- Given the publication date, we read the resource real-terms reduction + “next financial year” as 2027-28. The Welsh Government dates the capital + reduction explicitly by 2029-30. We keep the printed magnitude of the + percentages and state in the conditions that positive means a reduction. We keep + “Almost 10%” in `value_raw`, mark the upper endpoint 10 `range_high` and + keep the qualifier. +- The Welsh Government states no real-terms basis for the monetary table, + and we apply no deflation. We keep the percentage context rows in real + terms and invent no price year. Every row concerns Wales. +- We use `value_kind: point` for a single published figure, not a separately + identified central-model scenario, and use `central` on no row. +- We quote an exact page fragment or table cell in every row, containing + `value_raw` and fewer than 15 words. We read tables by row and then column. + +### Welsh figures we left out + +- The Welsh Government prints Resource 2029-30 as `N/A`, not a number and not zero. +- The Welsh Government restates OBR GDP growth of 1.2% in 2026 and 1.0% in + 2027 as macro context. That falls outside this assignment's + devolved-consequentials scope, and the Welsh Government asserts no own + estimate of a Budget measure there. +- The Welsh Government announces how it will implement the £100 + council-tax-band A-D payment, a policy parameter, but prints no + implementation date, reach, aggregate cost or estimated household gain. We + didn't attach it to the England-only Treasury line 1 key or fabricate an + impact row from it. +- We treat the £14 billion rail commitment and the £302m already allocated + over the spending review as existing commitments, not newly estimated + effects of AB2026; the statement doesn't isolate a Budget measure's + effect. +- We don't count dates, council tax band letters or rounding caveats as + estimates, and the Welsh Government prints no charts that need bar + readings. We use no inaccessible Welsh page or unavailable local copy as evidence. ## Scottish Government -The [news release](https://www.gov.scot/news/uk-budget-delivers-real-cut-to-scotlands-funding/) ("UK Budget delivers real cut to Scotland's funding", 28 October 2026) was fetched fresh on 2026-10-29 with the same User-Agent; HTTP 200, 27,662 bytes, sha256 `f3d0183e470f711b6821e4956e108c5d17210a392e6a6e471eccf297c807a6ac`. +The lane fetched the [news release](https://www.gov.scot/news/uk-budget-delivers-real-cut-to-scotlands-funding/) ("UK Budget delivers real cut to Scotland's funding", 28 October 2026) fresh on 2026-10-29 with the same User-Agent: HTTP 200, 27,662 bytes, sha256 `f3d0183e470f711b6821e4956e108c5d17210a392e6a6e471eccf297c807a6ac`. -- **Sign:** the primary prints unsigned magnitudes and describes reductions. Values retain those positive magnitudes, with `conditions.sign_convention: "positive = magnitude of reduction in Scottish Government funding"`. No minus sign is invented. The three million-pound amounts are multiplied by 1e6; the percentages remain percentages. -- **Period:** all rows use fiscal-year end year `2028`, `time_basis: fiscal_year`, and `conditions.fy: "2027-28"`. -- **Prices and comparison:** all rows say real terms, with an unspecified price base year. The primary does not explicitly name its comparison baseline. `conditions.price_basis` and `comparison_basis` preserve those limits. No cash-basis conversion is made. -- **Point/central:** the page gives one reported estimate per figure, not a named central model scenario, so `value_kind: point`. The money amounts retain the producer's “around” qualifier in `conditions.precision`. -- **Attribution:** all 5 are Scottish Government `own` estimates because the article reports its funding assessment without naming a separate original source. `arithmetic` follows the AB2025 family's funding convention and does not assert access to an unpublished model. None is an administrative outturn or count. -- **Baselines:** `baseline_policy: null`; no proposed baseline. These are context funding projections, not a comparison of a reform against a specified alternative policy. -- **Keys:** **`measure_key: null` — 5 rows**. No registered key is used, and **no proposed measure key** is introduced. The article does not isolate a set of Table 6.1 decisions or establish that its year-on-year real-terms funding assessment matches a registered package. `reform_hint: null` records that it is context. None of the 98 Table 6.1 lines is asserted to be the exact scope of these figures. +- **Sign.** The Scottish Government prints unsigned magnitudes and describes reductions. We keep those positive magnitudes, with `conditions.sign_convention: "positive = magnitude of reduction in Scottish Government funding"`, and invent no minus sign. We multiply the three million-pound amounts by 1e6 and keep the percentages as percentages. +- **Period.** On all rows we use fiscal-year end year `2028`, `time_basis: fiscal_year` and `conditions.fy: "2027-28"`. +- **Prices and comparison.** The Scottish Government gives every figure in real terms, with an unspecified price base year, and doesn't explicitly name its comparison baseline. We record those limits in `conditions.price_basis` and `comparison_basis`, and make no cash-basis conversion. +- **Point or central.** The Scottish Government reports one estimate per figure, not a named central model scenario, so we use `value_kind: point`. We keep its “around” qualifier on the money amounts in `conditions.precision`. +- **Attribution.** We mark all 5 as the Scottish Government's `own` estimates, because it reports its funding assessment without naming a separate original source. We use `arithmetic` to follow the AB2025 family's funding convention, without claiming access to an unpublished model. The Scottish Government reports no administrative outturn or count here. +- **Baselines.** We set `baseline_policy: null` and propose no baseline. We stage these funding projections as context: the Scottish Government doesn't compare a reform with a specified alternative policy. +- **Keys.** We give **5 rows `measure_key: null`**. We use no registered key and introduce **no proposed measure key**. The Scottish Government doesn't isolate a set of Table 6.1 decisions or show that its year-on-year real-terms funding assessment matches a registered package. We set `reform_hint: null` to record that we stage the rows as context. We assert none of the 98 Table 6.1 lines as the exact scope of these figures. -### Scottish rows not staged +### Scottish figures we left out -- Every FAI figure: the primary refused the required fresh fetch, and no local fallback was supplied. The family therefore has no read primary evidence for FAI's quantitative or qualitative claims. -- SFC: no qualifying publication dated 28–29 October was found. Its earlier forecasts and funding commentary are not recast as post-Budget figures. -- The Scottish Government's £100 household energy-support commitment: it is a payment-policy parameter, not a separate quantified aggregate or model estimate. The article does not specify Scottish eligibility, rollout timing or a fully defined Scottish implementation; the registered Table 6.1 line 1 describes England's council-tax rebate. No England measure key is attached to Scotland's unspecified implementation. -- The £300 energy-bill reduction refers to an earlier political promise, not an AB2026 estimate or a newly costed option. -- The repeated £650 million in the minister's quotation and repeated £100 commitment are not staged twice. -- Dates, earlier political reference years and qualitative references to welfare policies do not form numerical claims. No chart-only numbers, macro forecasts or third-party restatements were used. +- Every FAI figure: FAI's site refused the required fresh fetch, and nobody supplied a local fallback. So we read no primary evidence for FAI's quantitative or qualitative claims. +- SFC: we found no qualifying publication dated 28–29 October, and we don't recast its earlier forecasts and funding commentary as post-Budget figures. +- The Scottish Government's £100 household energy-support commitment: the Scottish Government sets a payment-policy parameter, not a separate quantified aggregate or model estimate. It specifies no Scottish eligibility, rollout timing or fully defined Scottish implementation, and on the registered Table 6.1 line 1 HM Treasury describes England's council-tax rebate. We attach no England measure key to Scotland's unspecified implementation. +- The Scottish Government's £300 energy-bill reduction refers to an earlier political promise, not an AB2026 estimate or a newly costed option. +- We don't stage the minister's repeated £650 million or the repeated £100 commitment twice. +- We don't treat dates, earlier political reference years or qualitative references to welfare policies as numerical claims. We used no chart-only numbers, macro forecasts or third-party restatements. ### Scottish Government primary issues -The response labels the figures as a consequence of the UK Budget but does not distinguish changes caused by newly announced decisions from the inherited funding path and inflation. It supplies no derivation, explicit comparison baseline or deflator year. Consequently the five context rows cannot be compared directly with a cash-terms Barnett-consequentials table. The two component cash magnitudes are consistent with the printed total; none was recomputed. +The Scottish Government attributes the figures to the UK Budget but doesn't separate the changes that newly announced decisions cause from the inherited funding path and inflation. It supplies no derivation, explicit comparison baseline or deflator year, so nobody can compare the five context rows directly with a cash-terms Barnett-consequentials table. The two component cash magnitudes agree with the printed total, and we recomputed none. -The HTML's JSON-LD `dateModified` is `2026-10-29T08:22:59Z`, while the visible publication time and `dc.date.modified` show 28 October 2026 15:45. No earlier local copy was provided, and the metadata do not establish that the numerical content changed. The manifest pins the sole fetched version and records both timestamps; there is no invented second version. +On its page the Scottish Government gives `2026-10-29T08:22:59Z` in JSON-LD `dateModified`, and 28 October 2026 15:45 in the visible publication time and `dc.date.modified`. Nobody provided an earlier local copy, and the metadata don't show that the numbers changed. We pin the sole fetched version in the manifest, record both timestamps there and invent no second version. ## Verification -- **4 November:** the contract checker reports 19 rows, 0 problems, and a re-run of `stage.py` reproduces both files byte for byte. -- **3 November:** the contract checker reports 19 rows, 0 problems; a re-run of `stage.py` reproduces `manifest.jsonl` and `claims_staged.jsonl.gz` byte for byte (COUNTS.json pins the gzip's sha256). The fixes relabel fields only, so no spot check was repeated; the lanes' checks below stand. -- **Welsh rows (29 October).** Ten rows selected with `random.Random(2302026).sample(range(14), 10)` -were checked against the saved, hash-verified HTML, reselecting the table -row/column or paragraph. Row numbers below are one-based. Every check +- On 4 November, the contract checker reported 19 rows, 0 problems, and when we re-ran `stage.py` it reproduced both files byte for byte. +- On 3 November, the contract checker reported 19 rows, 0 problems, and when we re-ran `stage.py` it reproduced `manifest.jsonl` and `claims_staged.jsonl.gz` byte for byte (COUNTS.json pins the gzip's sha256). We only relabelled fields, so we repeated no spot check, and the lanes' checks below stand. +- Welsh rows, 29 October: the lane selected ten rows with `random.Random(2302026).sample(range(14), 10)` +and checked them against the saved, hash-verified HTML, reselecting the table +row/column or paragraph. We give one-based row numbers below. Every check confirmed the number, sign/magnitude convention, unit and end-year period. | Row | Reopened locator | Printed value | Normalized value and period | @@ -156,12 +163,12 @@ confirmed the number, sign/magnitude convention, unit and end-year period. | 13 | First paragraph after table/rounding footnote, last sentence | 0.5% | 0.5 percent reduction; 2028 | | 14 | Second paragraph after table, second sentence | almost 10% | Upper endpoint 10 percent reduction; 2030 | -The headline and all remaining numeric cells were also read directly when -transcribing. Every row's quote contains its raw value. Scope review found -no assigned registry key. Since 3 November the Welsh package rows carry the -shared proposed key `ab2026__package_barnett_consequentials` (scope above), -and the context rows carry no proposal. -- **Scottish rows (29 October, in `uk_fai_sfc`).** Because the family contains fewer than 10 rows, all 5 were checked. A reproducibly shuffled order (`random.Random(20261029)`) gave rows **3, 2, 1, 4, 5** (now rows 17, 16, 15, 18, 19 of this family, one-based). The saved primary was re-opened at every locator; its hash was also checked against each row's publication hash. +The lane also read the headline and all remaining numeric cells directly +when transcribing, and every row's quote contains its raw value. Its scope +review found no assigned registry key. On 3 November we gave the Welsh +package rows the shared proposed key `ab2026__package_barnett_consequentials` +(scope above), which we registered in tranche 2, and the context rows no key. +- Scottish rows, 29 October, in `uk_fai_sfc`: we then had fewer than 10 rows in the family, so the lane checked all 5. It shuffled them reproducibly (`random.Random(20261029)`) and checked rows **3, 2, 1, 4, 5** in that order (now rows 17, 16, 15, 18, 19 of this family, one-based). It re-opened the saved primary at every locator and checked its hash against each row's publication hash. | Row | Primary locator | Verbatim figure | Checked normalization | Sign and period | | ---: | --- | --- | --- | --- | diff --git a/sources/harvest-uk-ab2026-2026-10-29/uk_hmt_costings/NOTES.md b/sources/harvest-uk-ab2026-2026-10-29/uk_hmt_costings/NOTES.md index 1a7800f55..dff01dd44 100644 --- a/sources/harvest-uk-ab2026-2026-10-29/uk_hmt_costings/NOTES.md +++ b/sources/harvest-uk-ab2026-2026-10-29/uk_hmt_costings/NOTES.md @@ -1,86 +1,93 @@ # uk_hmt_costings: HM Treasury, Budget 2026: Policy Costings (PU 3699) -**Working files.** `stage.py`, the file meant wherever these notes say `stage.py`, lives on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/hmt-costings/uk_hmt_costings/stage.py`; this branch carries only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz`. The repository carries no primary bytes: fetch the costings PDF from its `manifest.jsonl` URL and check its sha256. +We keep `stage.py` on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/hmt-costings/uk_hmt_costings/stage.py`. We commit only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz` to this branch, and no primary bytes to the repository: fetch the costings PDF from its `manifest.jsonl` URL and check its sha256. -This family dates from 2026-10-29 and follows README rules 1 to 10, including rule 8 ("stage estimates, not design"), rule 9 ("a component of one measure keeps that measure's key") and rule 10 (below, under Conventions). `stage.py` (on the staging branch; see 'Working files') holds the rows as data and rewrites `claims_staged.jsonl.gz` byte for byte. +We staged this family on 2026-10-29 and follow README rules 1 to 10. Those include rule 8 ("stage estimates, not design"), rule 9 ("a component of one measure keeps that measure's key") and rule 10 (below, under Conventions). We wrote the rows into `stage.py` (on the staging branch; see the paragraph above) as data, and it rewrites `claims_staged.jsonl.gz` byte for byte. -Source slug `hm_treasury` and source_model `hmt_costing_obr_certified`, as in the AB2025 family. The exception is the two tax gap rows, which use `administrative_data`. +We use source slug `hm_treasury` and source_model `hmt_costing_obr_certified`, as in the AB2025 family, except on the two tax gap rows, where we use `administrative_data`. -- **Rows:** 597. +- Rows: 597. - 576 Exchequer impact cells: 96 costing notes, each across six fiscal years from 2026-27 to 2031-32. - - 21 rows of estimates printed in the notes' text: + - 21 rows of estimates that HM Treasury prints in the notes' text: - behavioural parameters (8); - PSND and PSNFL impacts (7); - eligible counts (2); - the tax gap behind one tax base (2); - one tax base; - one financial-transaction impact. -- **Primaries:** 1 PDF. -- **Measure keys:** 76 registry keys on 573 rows. - - 416 rows cost a whole Table 6.1 line: `conditions.row_type: measure`, 68 keys. - - 157 rows cost a printed part of one line: `row_type: measure_component`, 8 lines' keys. - - 22 rows carry `measure_key: null` and one of 3 proposed keys, for measures with no Table 6.1 line or none this family can identify. 6 of them (p.108's receipts) are `measure_component`, so 163 rows in all are components. - - 2 rows (p.50's UK tax gap) are context: `measure_key: null` and no proposed key. +- Primaries: 1 PDF. +- Measure keys: 79 registry keys on 595 rows. + - We stage HM Treasury's costing of a whole Table 6.1 line on each of 416 rows: `conditions.row_type: measure`, 68 keys. + - We stage HM Treasury's costing of a printed part of one line on each of 157 rows: `row_type: measure_component`, 8 lines' keys. + - We key 22 rows for measures with no Table 6.1 line, or none we can identify, to 3 keys that we proposed and registered in tranche 2. We mark 6 of them (p.108's receipts) `measure_component`, for 163 component rows in all. + - We stage 2 rows (p.50's UK tax gap) as context: `measure_key: null` and no proposed key. ## Access recipe -- **Fetch:** `curl -sSL` with a browser User-Agent to https://assets.publishing.service.gov.uk/media/6ae1883fe4367e5c8b0467df/AB26_Policy_Costing.pdf on 2026-10-29 at 16:14 UTC. - - HTTP 200, `application/pdf`, 745,488 bytes, sha256 `f716efada8a4d0726ce12b0f97d3f0dac671c31b2d005ecf300a48ff309638bf`. - - The site showed no bot mitigation, so the fetch needed no fallback. - - The bytes equal the harvest's local copy downloaded on 28 October. The primary has not changed since Budget day, so README rule 7 does not apply. -- **Reading:** `pdftotext -layout`. - - 122 pages: the introduction on p.10, costing notes on pp.12-109 and Annex A on pp.110-121. +- We fetched https://assets.publishing.service.gov.uk/media/6ae1883fe4367e5c8b0467df/AB26_Policy_Costing.pdf with `curl -sSL` and a browser User-Agent on 2026-10-29 at 16:14 UTC. + - The server answered HTTP 200, `application/pdf`, 745,488 bytes, sha256 `f716efada8a4d0726ce12b0f97d3f0dac671c31b2d005ecf300a48ff309638bf`. + - We met no bot mitigation on the site, so we needed no fallback. + - The bytes equal the harvest's local copy downloaded on 28 October. HM Treasury hasn't changed the primary since Budget day, so README rule 7 doesn't apply. +- We read it with `pdftotext -layout`. + - HM Treasury prints 122 pages: the introduction on p.10, costing notes on pp.12-109 and Annex A on pp.110-121. - Two notes run onto a second page: EMI2 (pp.31-32, table on p.32) and Writing-down Allowances (pp.70-71, table on p.70). - - This file cites each note by its first page. A row's locator cites the page that prints the figure. -- **Where the data in `stage.py` comes from:** a scratch parser, not vendored, split the text into notes at each "Measure description" heading and read each note's sections. - - Every note prints exactly one "Exchequer impact (£m)" table: one row of six signed values for 2026-27 to 2031-32. - - I reviewed the parse against the page text. All 96 titles match the document's contents list. - - Prose joins a line-end hyphen without a space, except a suspended one: p.82's "18- to 24-year-old". -- **Re-check:** + - In this file we cite each note by its first page. In each row's locator we cite the page where HM Treasury prints the figure. +- To get the data in `stage.py`, we split the text into notes at each "Measure description" heading with a scratch parser, which we didn't vendor, and read each note's sections. + - HM Treasury prints exactly one "Exchequer impact (£m)" table in every note: one row of six signed values for 2026-27 to 2031-32. + - We reviewed the parse against the page text. All 96 titles match the document's contents list. + - We join a line-end hyphen in prose without a space, except a suspended one: p.82's "18- to 24-year-old". +- To re-check: - `python3 stage.py` rewrites the gzip. - - `python3 stage.py --verify ` writes nothing. It checks that the gzip on disk equals a fresh build and that the PDF has the pinned sha256. For every row it then checks the page its locator cites: + - `python3 stage.py --verify ` writes nothing. It checks that the gzip on disk equals a fresh build and that the PDF has the pinned sha256. For every row it then checks the page we cite in its locator: - **table rows:** the heading, the six years, the six values in order and the row's value in its year's column. The row label must equal the page's label exactly, and any footnote must appear. - - **text rows:** the quote, verbatim, with `value_raw` inside it. `value` must equal `value_raw`'s number at the scale and sign the `normalization` names. + - **text rows:** the quote, verbatim, with `value_raw` inside it. `value` must equal `value_raw`'s number at the scale and sign we name in `normalization`. - **every keyed row:** the measure's title, on the note's first page. - - All 597 rows pass. An in-memory mutation test confirms the check catches each of: a wrong table value; swapped columns; a wrong or shortened label (p.82 without "[AME savings only]"); a text figure off by sign, scale or value; and a wrong title. -- **Table quotes.** A table row's `quote` joins the table's cells in reading order: heading | years | label and six values | footnote. p.96 prints its label broken across two lines as "Exchequ / er impact"; the quote and `table_row` write it "Exchequer impact". + - All 597 rows pass. In an in-memory mutation test, the check caught each of: a wrong table value; swapped columns; a wrong or shortened label (p.82 without "[AME savings only]"); a text figure off by sign, scale or value; and a wrong title. +- In a table row's `quote` we join the table's cells in reading order: heading | years | label and six values | footnote. On p.96 HM Treasury breaks its label across two lines as "Exchequ / er impact"; in the quote and `table_row` we write it "Exchequer impact". -## What the family stages +## What we staged ### Exchequer impact tables (576 rows) -- **One row per note and fiscal year:** - - `metric: revenue_change` and `unit_concept: gbp`. - - `value` is `value_raw` (£m, with the sign and trailing "m" as printed) × 1e6. -- **Every table is `scoring_method: final`.** - - No AB2026 note prints separate static and post-behavioural tables; in AB2025, 8 notes did. - - No note prints a behavioural adjustment in pounds; AB2025's thresholds note printed £180m. - - So the family has no static, post-behavioural or behavioural-effect legs to stage. Several notes mention a static costing, and p.108 a "pre-behavioural costing", but none prints a figure for it. -- **`conditions.behavioural_adjustment`** classifies the words of each note's costing section. The pages listed are each note's first page. - - **"included (see note)" on 56 notes:** pp.12, 13, 14, 16, 18, 19, 23, 24, 25, 26, 27, 29, 30, 31, 36, 38, 39, 40, 43, 44, 45, 46, 47, 48, 49, 51, 52, 53, 60, 62, 65, 66, 67, 68, 70, 72, 73, 74, 75, 77, 78, 79, 80, 81, 82, 85, 86, 87, 90, 95, 100, 102, 103, 104, 107 and 108. - - p.90 adds its sentence: the Infected Blood costing keeps the scheme's cost model, whose "behavioural assumptions" the changes "do not materially alter". +- We stage one row per note and fiscal year: + - with `metric: revenue_change` and `unit_concept: gbp`; + - with `value` = `value_raw` (£m, with the sign and trailing "m" as printed) × 1e6. +- We mark every table `scoring_method: final`. + - HM Treasury prints no separate static and post-behavioural tables in any AB2026 note; in AB2025, 8 notes did. + - HM Treasury prints no behavioural adjustment in pounds in any note; in AB2025's thresholds note it printed £180m. + - So we had no static, post-behavioural or behavioural-effect legs to stage. HM Treasury mentions a static costing in several notes and a "pre-behavioural costing" on p.108, but prints no figure for any of them. +- In `conditions.behavioural_adjustment` we classify the words of each note's costing section. We list each note by its first page. + - **"included (see note)" on 56 notes:** + - pp.12, 13, 14, 16, 18, 19, 23, 24, 25 and 26; + - pp.27, 29, 30, 31, 36, 38, 39, 40, 43 and 44; + - pp.45, 46, 47, 48, 49, 51, 52, 53, 60 and 62; + - pp.65, 66, 67, 68, 70, 72, 73, 74, 75 and 77; + - pp.78, 79, 80, 81, 82, 85, 86, 87, 90 and 95; + - pp.100, 102, 103, 104, 107 and 108. + + For p.90 we add HM Treasury's sentence: it keeps the scheme's cost model for the Infected Blood costing, and says the changes "do not materially alter" the model's "behavioural assumptions". - **"none: ''" on 18 notes:** pp.17, 35, 37, 42, 50, 54, 55, 56, 58, 59, 61, 63, 64, 91, 94, 101, 106 and 109. - **"not stated (the note's costing section does not use the word 'behaviour' or 'behavioural')" on 22 notes:** pp.15, 20, 21, 22, 28, 33, 34, 41, 57, 69, 76, 83, 84, 88, 89, 92, 93, 96, 97, 98, 99 and 105. - - This class is lexical and reads the costing section only; AB2025 used only the first two classes. - - Six of these notes name behaviour in their "Areas of uncertainty" instead: + - We assign this class by the words alone and read only the costing section; in AB2025 we used only the first two classes. + - In six of these notes HM Treasury names behaviour in "Areas of uncertainty" instead: - pp.20, 21, 22, 69 and 89: "the impact of behavioural responses"; - p.93: "the behavioural response to the rebate scheme". - - Others describe a response in other words: - - p.15 assumes 89% take-up; - - p.28 puts 25% of income into reserves; - - p.76 "includes an assumed timing effect"; - - p.84 assumes the zone facilitates borrowing; - - p.88 says self-assessment "may affect how taxpayers interact with the tax system"; - - p.99 adjusts "for expected utilisation of the cap". -- **`conditions.static_impact`:** three notes say the measure has no static impact, so the whole costing is behavioural. + - In others HM Treasury describes a response in other words: + - p.15: assumes 89% take-up; + - p.28: puts 25% of income into reserves; + - p.76: "includes an assumed timing effect"; + - p.84: assumes the zone facilitates borrowing; + - p.88: says self-assessment "may affect how taxpayers interact with the tax system"; + - p.99: adjusts "for expected utilisation of the cap". +- `conditions.static_impact`: in three notes HM Treasury says the measure has no static impact, so the whole costing comes from behaviour. - p.19 (Cash ISAs) and p.43 (E-invoicing): "The costing has no static impact on taxpayer liabilities." - p.44 (Employment expenses): "The measure has no static impact on taxpayer liabilities." -- **`conditions.table_row`:** "Exchequer impact" on 95 notes; "Exchequer impact [AME savings only]" on p.82. -- **`conditions.table_footnote`:** +- `conditions.table_row`: "Exchequer impact" on 95 notes; "Exchequer impact [AME savings only]" on p.82. +- `conditions.table_footnote`: - p.94: "*Note: This costing also includes delivery costs." - p.108: "Note: this is offset by increased DEL spending so is fiscally neutral." -- **`conditions.fiscally_neutral`:** pp.108 and 109, each with the note's own sentence. -- **`note`:** the note's base section, then its costing section, then "Areas of uncertainty", verbatim and joined by " || ". Component and off-table notes have a mapping paragraph in front (see "Measure keys"). +- `conditions.fiscally_neutral`: pp.108 and 109, each with the note's own sentence. +- In `note` we copy the note's base section, then its costing section, then "Areas of uncertainty", verbatim and joined by " || ". For component and off-table notes we put a mapping paragraph in front (see "Measure keys"). ### Estimates from the notes' text (21 rows) @@ -91,69 +98,69 @@ Source slug `hm_treasury` and source_model `hmt_costing_obr_certified`, as in th | 28 | "25% of the additional income above the baseline is placed into MSAs’ reserves" | `behavioural_parameter`, percent | 2027-28 | `ab2026__business_rates_retention_mayoral_authorities` | | 33 | "public sector net debt (PSND) (approx. £386m) and public sector net financial liabilities (PSNFL) (approx. £375m)" | `psnd_impact` -386m, `psnfl_impact` -375m | to 2031-32, cumulative | line 29, component p.33 | | 34 | "a cost of approximately £775m" (PSND); "add an additional cost of £68m" (PSNFL fair value) | `psnd_impact` -775m, `psnfl_fair_value_impact` -68m | to 2031-32, cumulative | line 29, component p.34 | -| 41 | "a cost of approximately £32m" (PSND); "the value of the portfolio increases by approximately £86m (total PSNFL benefit of £54m)" | `psnd_impact` -32m, `psnfl_fair_value_impact` +86m, `psnfl_impact` +54m | to 2031-32, cumulative | proposed, p.41 | +| 41 | "a cost of approximately £32m" (PSND); "the value of the portfolio increases by approximately £86m (total PSNFL benefit of £54m)" | `psnd_impact` -32m, `psnfl_fair_value_impact` +86m, `psnfl_impact` +54m | to 2031-32, cumulative | `ab2026__british_business_bank_defence_commitments_accelerated` | | 50 | "6.4% of total theoretical tax liabilities, or £59.2 billion in absolute terms" | `tax_gap`, percent and gbp | 2024-25, outturn | context (`measure_key: null`) | | 58 | "an impact of +£130m estimated in 2029-30 in cash terms" | `financial_transaction_cash_impact` | 2029-30 | line 47, component p.58 | | 67 | "An elasticity of -0.95 is used, with 25% of the reduction in stakes then assumed to substitute to Remote Gaming Duty" | `behavioural_parameter`, elasticity and percent | 2027-28 | line 56, component p.67 | | 84 | "£20 million of additional capital borrowing annually from 2028-29 to 2031-32" | `behavioural_parameter`, gbp, one row per year | 2028-29 to 2031-32 | `ab2026__leeds_city_fund` | -| 108 | "approximately 4.9 million eligible properties are in scope" | `tax_base`, properties | 2026, point in time | proposed, p.108 | +| 108 | "approximately 4.9 million eligible properties are in scope" | `tax_base`, properties | 2026, point in time | `ab2026__private_landlord_register_fee_65`, its tax base | -- **Labels.** Each text row names its quantity in `conditions.statistic`, or `conditions.parameter` for a behavioural parameter. Rule 9 reserves `conditions.component` for the part of the measure. -- **Per-year rows.** p.84's sentence gives one figure for each of four named years, so the family stages one row per year, each quoting the whole sentence. The family sums and splits nothing. -- **Year of a parameter.** The note gives no year for p.15's take-up, p.28's reserves share or p.67's two parameters. As in the AB2025 family, `fy` is the measure's first year, and the row's `note` says so. -- **Present-tense counts** (p.15 "There are an estimated", p.108 "The current estimate") print no year. They carry `period: 2026` and `time_basis: point_in_time`, with `basis: outturn`. +- **Labels.** We name each text row's quantity in `conditions.statistic`, or in `conditions.parameter` for a behavioural parameter. Under rule 9 we use `conditions.component` only for the part of the measure. +- **Per-year rows.** In p.84's sentence HM Treasury gives one figure for each of four named years, so we stage one row per year, each quoting the whole sentence. We sum and split nothing. +- **Year of a parameter.** HM Treasury gives no year for p.15's take-up, p.28's reserves share or p.67's two parameters. As in the AB2025 family, we set `fy` to the measure's first year and say so in the row's `note`. +- **Present-tense counts.** HM Treasury prints no year for p.15's "There are an estimated" or p.108's "The current estimate". We give them `period: 2026` and `time_basis: point_in_time`, with `basis: outturn`. - **Metric choices that could go either way:** - - p.15's population is the people a new benefit would cover, so it takes the registered `eligible_count`. - - p.108's 4.9 million is the base the fee falls on, so it takes `tax_base`, as AB2025's 165,000 HVCTS properties did. - - p.84's £20 million is the council's estimated response to the measure ("Additional business rates income is assumed to facilitate borrowing", and "none is assumed without the zone"). So it is a `behavioural_parameter` in pounds, as the AB2025 uk_obr_tables family staged costing assumptions (its £1,849 row), not a financing remit. + - On p.15 HM Treasury counts the people a new benefit would cover, so we use the registered `eligible_count`. + - On p.108 HM Treasury counts the base the fee falls on (4.9 million), so we use `tax_base`, as we did for AB2025's 165,000 HVCTS properties. + - On p.84 HM Treasury gives £20 million as the council's estimated response to the measure ("Additional business rates income is assumed to facilitate borrowing", and "none is assumed without the zone"). So we stage it as a `behavioural_parameter` in pounds, not a financing remit, as we staged costing assumptions in the AB2025 uk_obr_tables family (its £1,849 row). - **PSND and PSNFL figures.** - - They sum over "the period to 2031-32", so they carry `value_kind: cumulative`, `period: 2032`, `conditions.horizon` and no `fy`. - - The document prints them unsigned, with words. The family gives all seven one convention, matching the Exchequer rows: positive = lower PSND or PSNFL (an improvement); negative = higher (a cost). - - `normalization` and `sign_convention` quote the words each sign comes from: "net cost", "a cost of" and "add an additional cost of" are negative; "benefit" and "the value of the portfolio increases by" are positive. - - The £68m (p.34) and the £86m (p.41) are fair-value movements: increments to PSNFL beyond PSND, not PSNFL totals. They take their own proposal, `psnfl_fair_value_impact`. + - HM Treasury sums them over "the period to 2031-32", so we give them `value_kind: cumulative`, `period: 2032`, `conditions.horizon` and no `fy`. + - HM Treasury prints them unsigned, with words. We give all seven one convention, matching the Exchequer rows: positive = lower PSND or PSNFL (an improvement); negative = higher (a cost). + - In `normalization` and `sign_convention` we quote the words each sign comes from. We read "net cost", "a cost of" and "add an additional cost of" as negative, and "benefit" and "the value of the portfolio increases by" as positive. + - HM Treasury gives the £68m (p.34) and the £86m (p.41) as fair-value movements: increments to PSNFL beyond PSND, not PSNFL totals. We propose `psnfl_fair_value_impact` for them. - p.41's figures agree: -32 + 86 = +54. - - p.34 prints no PSNFL total, so the family stages none. + - HM Treasury prints no PSNFL total on p.34, so we stage none. - **The tax gap.** - - p.50 prints HMRC's tax gap for 2024 to 2025 in its tax base section, as the base's starting point: "The tax base for this costing is the total tax gap less the additional tax revenues accounted for by previously announced Closing the Tax Gap (CTG) initiatives." - - Rule 8 counts a tax base as an estimate, but the total tax gap is a context figure, not a part of the measure, so the two rows carry `measure_key: null` (AB2025 rule 3). - - They are an official HMRC statistic repeated by HM Treasury, so they carry `attribution: restated`, `benchmark_class: administrative_fact`, `source_model: administrative_data` and `basis: outturn`, with the uk_rf_ab2025 proposal spelling `tax_gap`. - - The note does not cite the HMRC publication. - -## Conventions (load-bearing) - -- **Sign.** The document prints no legend. - - `conditions.sign_convention` on every Exchequer row reads "positive = yield to the Exchequer (reduces borrowing); negative = cost to the Exchequer". The full text also gives the evidence: - - p.38 says its figures "represent net income from developer fees" and prints +50m to +95m from 2027-28, after -15m in 2026-27. - - The Bank Surcharge rise prints +, and the VAT cut on domestic electricity prints -. - - Table 6.1 prints the same values with the same signs. - - Values are NOT re-signed. - - Text rows carry their own `sign_convention`: as printed for p.58's "+£130m", and for the PSND and PSNFL rows as described above. -- **Rule 10 (borrowing metric).** pp.34, 35 and 41 say their tables' figures "impact public sector net borrowing (PSNB)". They are single measures' Exchequer impacts, which Table 6.1 frames the same way (red book para 6.1), and README rule 10 keeps single measures on `revenue_change` unless the producer states the Budget's effect on borrowing. So rows 134–145 and 179–184 keep `revenue_change`; p.35 equals line 30. No row in this family takes `decisions_effect_on_borrowing`. -- **period** is the FY end year (2026-27 → 2027), with `conditions.fy` holding the label. -- **geography** is "UK" on every row, because these are Exchequer impacts. `conditions.measure_scope` names the territory where the note or its title says the measure applies: + - On p.50 HM Treasury prints HMRC's tax gap for 2024 to 2025 in its tax base section, as the base's starting point: "The tax base for this costing is the total tax gap less the additional tax revenues accounted for by previously announced Closing the Tax Gap (CTG) initiatives." + - Under rule 8 we count a tax base as an estimate, but we treat the total tax gap as a context figure, not a part of the measure. So we give the two rows `measure_key: null` (AB2025 rule 3). + - HM Treasury repeats an official HMRC statistic here, so we mark the rows `attribution: restated`, `benchmark_class: administrative_fact`, `source_model: administrative_data` and `basis: outturn`, with the uk_rf_ab2025 proposal spelling `tax_gap`. + - HM Treasury doesn't cite the HMRC publication. + +## Conventions the rows depend on + +- **Sign.** HM Treasury prints no legend. + - On every Exchequer row we set `conditions.sign_convention` to "positive = yield to the Exchequer (reduces borrowing); negative = cost to the Exchequer". HM Treasury's full text supports it: + - on p.38 it says its figures "represent net income from developer fees" and prints +50m to +95m from 2027-28, after -15m in 2026-27; + - it prints the Bank Surcharge rise as + and the VAT cut on domestic electricity as -; + - it prints the same values with the same signs in Table 6.1. + - We didn't re-sign any value. + - We give text rows their own `sign_convention`: as printed for p.58's "+£130m", and for the PSND and PSNFL rows as described above. +- **Rule 10 (borrowing metric).** On pp.34, 35 and 41 HM Treasury says its tables' figures "impact public sector net borrowing (PSNB)". It gives them as single measures' Exchequer impacts, framed as in Table 6.1 (red book para 6.1). Under README rule 10 we keep single measures on `revenue_change` unless the producer states the Budget's effect on borrowing. So we keep `revenue_change` on rows 134–145 and 179–184; p.35 equals line 30. We give no row in this family `decisions_effect_on_borrowing`. +- **period.** We use the FY end year (2026-27 → 2027) and put the label in `conditions.fy`. +- **geography.** We set "UK" on every row, because HM Treasury reports Exchequer impacts. Where HM Treasury says in the note or its title where the measure applies, we name the territory in `conditions.measure_scope`: - England: pp.20, 21, 22, 36, 69, 96, 102 and 108. - Great Britain: pp.15 and 59. - England and Northern Ireland: p.99. - - pp.91 and 94 name England only for their cost bases ("English borrowers", "English undergraduate student loan outlay"), so they carry no `measure_scope`. -- **baseline_policy** is null throughout: current law at scoring, including the pre-announced indexation in Annex A. There is no `proposed_baseline`. -- **impact_channel** comes from the Table 6.1 line's Tax or Spend head in the registry: + - On pp.91 and 94 HM Treasury names England only for the cost bases ("English borrowers", "English undergraduate student loan outlay"), so we set no `measure_scope`. +- **baseline_policy.** We leave it null throughout, for current law at scoring, including the pre-announced indexation in Annex A. We propose no `proposed_baseline`. +- **impact_channel.** We take it from the Table 6.1 line's Tax or Spend head in the registry: - `tax_measure` or `spending_measure`; - `not_on_table_6_1` for pp.108-109; - `not_identified_on_table_6_1` for p.41. -- **national_accounts_basis**: "All costings are presented on a National Accounts basis (p.10)" on every Exchequer row. -- **parse_confidence** is "high" on every row. The document prints every value as text, and `--verify` finds each on its page. +- **national_accounts_basis.** We set "All costings are presented on a National Accounts basis (p.10)" on every Exchequer row. +- **parse_confidence.** We set "high" on every row. HM Treasury prints every value as text, and `--verify` finds each on its page. ## Measure keys -The document orders its costing notes by Table 6.1 line. Mapping each note to its line gives a non-decreasing sequence from line 2 to line 98, with the two Table 6.1 exclusions (pp.108-109) at the end. That ordering, together with titles and Table 6.1's footnotes, settles each note's line. +HM Treasury orders its costing notes by Table 6.1 line. When we map each note to its line, we get a non-decreasing sequence from line 2 to line 98, with the two Table 6.1 exclusions (pp.108-109) at the end. We settle each note's line from that ordering, the titles and Table 6.1's footnotes. -- **Whole lines (68 keys, 416 rows, `row_type: measure`).** These cover every note that is the only costing for its Table 6.1 line and costs all of it. +- **Whole lines (68 keys, 416 rows, `row_type: measure`).** We use these for every note in which HM Treasury gives the only costing for its Table 6.1 line and costs all of it. - For 64 notes the title equals the line's title exactly. - Four differ only by Table 6.1's parenthetical or the year: p.35 (line 30), p.38 (line 37), p.42 (line 44) and p.77 (line 66). -- **Components (8 lines, 25 notes, 157 rows, `row_type: measure_component`).** Under README rule 9, a note that costs a printed part of one Table 6.1 line carries that line's key. - - `conditions.component` names the part: the note's title, or for p.82 its fiscal head. - - `conditions.table_61_line` gives the line. - - The harvest never sums a component with the line's total. +- **Components (8 lines, 25 notes, 157 rows, `row_type: measure_component`).** Under README rule 9, we key a note that costs a printed part of one Table 6.1 line to that line. + - We name the part in `conditions.component`: the note's title, or for p.82 its fiscal head. + - We give the line in `conditions.table_61_line`. + - We never sum a component with the line's total in the harvest. | line | key | component notes (first page) | |---|---|---| @@ -166,103 +173,103 @@ The document orders its costing notes by Table 6.1 line. Mapping each note to it | 84 | `ab2026__waste_crime_landfill_tax_clearances` | p.92 six priority site clearances; p.93 local authority pilot | | 85 | `ab2026__further_priorities_nhs_and_vawg` | p.94 Erasmus+ (by position; see below) | -- **Proposed keys (3 keys, 22 rows; `row_type: measure`, except p.108's six receipts rows, which are `measure_component`).** Rule 9 keeps proposed keys "for scopes that cross Table 6.1 lines, or have none". The registry adds them before ingest; the table gives each key's scope as the producer's own words. +- **Keys with no Table 6.1 line.** We use 3 keys on 22 rows, all `row_type: measure` except p.108's six receipts rows, which we mark `measure_component`. Under rule 9 we proposed keys only "for scopes that cross Table 6.1 lines, or have none", and in tranche 2 we registered these three. In the table we give each key's scope in the producer's own words. -| proposed key | rows | page | scope | +| key | rows | page | scope | |---|--:|---|---| -| `ab2026__british_business_bank_defence_commitments_accelerated` | 9 | 41 | No identifiable Table 6.1 line (see below). "British Business Bank: Accelerate £450m of commitments to support defence-related and dual-use businesses" By Table 6.1's footnote 2 and its £10m bar, this costing is very probably inside a Table 6.1 line, most plausibly line 41 (lines 29 and 44 are fully accounted for by their own notes); tranche 2 must never add it to Table 6.1 lines or totals. | -| `ab2026__private_landlord_register_fee_65` | 7 | 108 | Not on Table 6.1 (footnote 2). The six Exchequer rows are the fee's receipts, a component (`component: "Receipts (fee income; offset by increased DEL spending, fiscally neutral)"`): p.108's note says "this is offset by increased DEL spending so is fiscally neutral", and the OBR stages the measure's total as 0 (review A, C-1). The seventh row is the 4.9 million tax base. "Renters' Rights: Roll out a £65 fee for the 'Register your rental property' service for private landlords and private properties in England from 15 December 2026" | +| `ab2026__british_business_bank_defence_commitments_accelerated` | 9 | 41 | No identifiable Table 6.1 line (see below). "British Business Bank: Accelerate £450m of commitments to support defence-related and dual-use businesses" From Table 6.1's footnote 2 and its £10m bar, we judge that HM Treasury very probably counts this costing inside a Table 6.1 line, most plausibly line 41 (the notes for lines 29 and 44 fully account for those lines). At ingest we score its six Exchequer rows (179–184) on this key alone and sum no row into a Table 6.1 line or total. | +| `ab2026__private_landlord_register_fee_65` | 7 | 108 | Not on Table 6.1 (footnote 2). We stage the fee's receipts, a component (`component: "Receipts (fee income; offset by increased DEL spending, fiscally neutral)"`), on the six Exchequer rows. On p.108 HM Treasury says "this is offset by increased DEL spending so is fiscally neutral", and we stage the OBR's total for the measure as 0 (review A, C-1). We stage the 4.9 million tax base on the seventh. "Renters' Rights: Roll out a £65 fee for the 'Register your rental property' service for private landlords and private properties in England from 15 December 2026" | | `ab2026__pwlb_debt_relief_councils_severe_distress` | 6 | 109 | Not on Table 6.1 (footnote 2). "Public Works Loan Board (PWLB) Debt Relief for Councils in Severe Financial Distress" | -How I settled each line: +How we settled each line: -- **Lines 45 and 47.** Table 6.1's footnote 4 on both lines reads "Full costings for the measures in this package are set out in the Budget 2026 Policy Costings document." - - The ten notes on pp.43-52 sit between line 44's note (p.42) and line 46's (p.53). - - The five notes on pp.54-58 sit between line 46's and line 48's (p.59). -- **Lines 29, 56, 77 and 84.** The line's title names both notes' measures. -- **Line 72 (p.82).** The note's title equals the line's, but its figure is one fiscal head of the line: - - its table prints the AME savings only; - - its text costs the Youth Guarantee's expansion to 22- to 24-year-olds and never mentions the apprenticeship support the line also names. +- **Lines 45 and 47.** On both lines HM Treasury writes in Table 6.1's footnote 4: "Full costings for the measures in this package are set out in the Budget 2026 Policy Costings document." + - HM Treasury puts the ten notes on pp.43-52 between line 44's note (p.42) and line 46's (p.53). + - It puts the five notes on pp.54-58 between line 46's and line 48's (p.59). +- **Lines 29, 56, 77 and 84.** In the line's title HM Treasury names both notes' measures. +- **Line 72 (p.82).** The note's title equals the line's, but its figure covers one fiscal head of the line: + - HM Treasury prints the AME savings only in its table; + - in its text HM Treasury costs the Youth Guarantee's expansion to 22- to 24-year-olds and never mentions the apprenticeship support it also names on the line. - Table 6.1 prints -290 and -375 where the note prints +25 and +35 (2027-28 and 2028-29). -- **Line 85 (p.94, Erasmus+).** No Table 6.1 title names Erasmus+. - - The note sits where line 85 falls: "Further Priorities: Fund key priorities, including pressures facing the NHS and delivery of the Violence against Women and Girls Strategy". + HM Treasury prints -290 and -375 in Table 6.1 and +25 and +35 in the note (2027-28 and 2028-29). +- **Line 85 (p.94, Erasmus+).** HM Treasury names Erasmus+ in no Table 6.1 title. + - HM Treasury puts the note where line 85 falls: "Further Priorities: Fund key priorities, including pressures facing the NHS and delivery of the Violence against Women and Girls Strategy". - Line 85's 2029-30 to 2031-32 figures (-5, 0, -15) equal the note's. - - Table 6.1's footnote 2 leaves out only two measures, neither of them Erasmus+, and the note's figures exceed the footnote's £10m bar in four of the six years. - - `conditions.document_position` records the position. -- **p.41.** Position narrows the note to lines 40-43, the national-security block. - - Table 6.1 prints lines 40-43 under its Spend head and prints no finer fiscal head, so nothing printed singles out one line. (The workbook's hidden column D, "Fiscal Head check (hide)", is not printed, and this family no longer reasons from it; review A, T-3.) - - But line 41's title, "Defence Investment Plan: Reallocate capital spending to fully fund the Defence Investment Plan", does not name the British Business Bank, and no figure ties the two. - - So the row carries a proposed key and `document_position`, not line 41's key. By Table 6.1's footnote 2 and its £10m bar, this costing is very probably inside a Table 6.1 line, most plausibly line 41 (lines 29 and 44 are fully accounted for by their own notes); tranche 2 must never add it to Table 6.1 lines or totals. -- **pp.108-109.** Table 6.1's footnote 2 reads "Measures with no net increase to borrowing are not included, these are debt relief for councils in severe financial distress through the Public Works Loan Board and a landlord registration fee for the Private Rented Sector." + - In Table 6.1's footnote 2 HM Treasury leaves out only two measures, neither of them Erasmus+, and the note's figures exceed the footnote's £10m bar in four of the six years. + - We record the position in `conditions.document_position`. +- **p.41.** By position, the note falls within lines 40-43, the national-security block. + - HM Treasury prints lines 40-43 under Table 6.1's Spend head and prints no finer fiscal head, so nothing printed singles out one line. (HM Treasury doesn't print the workbook's hidden column D, "Fiscal Head check (hide)", and since review A's T-3 we no longer reason from it.) + - But line 41's title, "Defence Investment Plan: Reallocate capital spending to fully fund the Defence Investment Plan", doesn't name the British Business Bank, and no figure ties the two. + - So we give the row its own key and `document_position`, not line 41's key. From Table 6.1's footnote 2 and its £10m bar, we judge that HM Treasury very probably counts this costing inside a Table 6.1 line, most plausibly line 41 (the notes for lines 29 and 44 fully account for those lines). At ingest we score its six Exchequer rows (179–184) on this key alone and sum no row into a Table 6.1 line or total. +- **pp.108-109.** HM Treasury explains the exclusions in Table 6.1's footnote 2. It writes: "Measures with no net increase to borrowing are not included, these are debt relief for councils in severe financial distress through the Public Works Loan Board and a landlord registration fee for the Private Rented Sector." -No `row_type: measure` row carries a registry key whose title covers more or less than the row's figure. +We put a registry key on a `row_type: measure` row only when the key's title covers exactly the row's figure. ## Cross-check against Table 6.1 (uk_hmt_tables) - **All 68 whole-line keys** equal Table 6.1 in all six years. -- **Component lines** (Table 6.1 against the components' sum, 2026-27 to 2031-32; computed for this check only, never staged; reported, not fixed): +- **Component lines.** We compared Table 6.1 with the components' sum, 2026-27 to 2031-32, for this check only; we never stage the sums, and we report the gaps without fixing them. - **Line 29 (pp.33+34):** equal. - - **Line 56 (pp.67+68):** the components sum to 140 in 2031-32 against Table 6.1's 145. The other years are equal. - - **Line 77 (pp.85+86):** the components sum to -170 in 2026-27 against Table 6.1's -175. The other years are equal. + - **Line 56 (pp.67+68):** the components sum to 140 in 2031-32 against Table 6.1's 145. The other years match. + - **Line 77 (pp.85+86):** the components sum to -170 in 2026-27 against Table 6.1's -175. The other years match. - **Line 84 (pp.92+93):** the components sum to -4, -3, -3 in 2027-28 to 2029-30 against Table 6.1's -5, -5, -5. - - These three gaps are each £5m or less, and every Table 6.1 cell is a multiple of £5m. That fits Table 6.1 rounding each line from unrounded components, but neither document says so. - - **Line 72 (p.82):** the AME component prints +0, +25, +35, +0, +0, +0 against Table 6.1's 0, -290, -375, 0, 0, 0. - - **Line 45 (pp.43-52):** the ten components sum to 110, 290, 240, 2,140, 2,635, 3,360 against Table 6.1's 10, 100, 115, 1,980, 2,485, 3,160. Table 6.1 is lower by 100, 190, 125, 160, 150 and 200. - - **Line 47 (pp.54-58):** the five components sum to -10, 10, 125, 65, 145, 235 against Table 6.1's 195, -125, 60, 75, 155, 240. The differences are +205, -135, -65, +10, +10 and +5. - - **Line 85 (p.94):** equal from 2029-30 to 2031-32. Table 6.1 prints larger costs from 2026-27 to 2028-29: -355, -175, -140 against the note's -15, +70, +45. + - Each of these three gaps comes to £5m or less, and HM Treasury prints every Table 6.1 cell as a multiple of £5m. That fits HM Treasury rounding each Table 6.1 line from unrounded components, but HM Treasury says so in neither document. + - **Line 72 (p.82):** HM Treasury prints +0, +25, +35, +0, +0, +0 for the AME component against Table 6.1's 0, -290, -375, 0, 0, 0. + - **Line 45 (pp.43-52):** the ten components sum to 110, 290, 240, 2,140, 2,635, 3,360 against Table 6.1's 10, 100, 115, 1,980, 2,485, 3,160. Table 6.1 falls short by 100, 190, 125, 160, 150 and 200. + - **Line 47 (pp.54-58):** the five components sum to -10, 10, 125, 65, 145, 235 against Table 6.1's 195, -125, 60, 75, 155, 240. The differences come to +205, -135, -65, +10, +10 and +5. + - **Line 85 (p.94):** equal from 2029-30 to 2031-32. HM Treasury prints larger costs in Table 6.1 from 2026-27 to 2028-29: -355, -175, -140 against the note's -15, +70, +45. - **Table 6.2:** p.58's "+£130m estimated in 2029-30 in cash terms" equals Table 6.2 line 9 (DWP Fraud and Error) for 2029-30. -- **Lines without a costing note:** 22 Table 6.1 lines have none: 1, 4, 9, 20, 22, 23, 24, 25, 31, 32, 34, 35, 40, 41, 42, 43, 63, 64, 67, 73, 74 and 81. - - The introduction limits the document to "tax and annually managed expenditure (AME) policy decisions". - - Table 6.1 prints all 22 under its Spend head. It prints no finer fiscal head, so whether each is departmental spending (outside the document's scope) is not printed; no note costs them. +- **Lines without a costing note:** HM Treasury gives 22 Table 6.1 lines no note: 1, 4, 9, 20, 22, 23, 24, 25, 31, 32, 34, 35, 40, 41, 42, 43, 63, 64, 67, 73, 74 and 81. + - In the introduction HM Treasury limits the document to "tax and annually managed expenditure (AME) policy decisions". + - HM Treasury prints all 22 under Table 6.1's Spend head and no finer fiscal head, so it doesn't say whether each counts as departmental spending (outside the document's scope); HM Treasury costs them in no note. ## Things in the primary that look wrong or contradictory (reported, not fixed) -- **The £10m threshold.** The introduction says "Measures that were certified by the OBR as below the negligible threshold of £10 million per annum in every year of the forecast period are not considered in this document." Five notes print costings below £10m in every year: +- **The £10m threshold.** In the introduction HM Treasury says "Measures that were certified by the OBR as below the negligible threshold of £10 million per annum in every year of the forecast period are not considered in this document." Yet HM Treasury prints costings below £10m in every year in five notes: - p.59 (+0m, then +5m a year); - p.92 (-2m, -2m, -3m); - p.93 (-2m, -1m); - - p.33 and p.109 (+0m in all six years; p.33's PSND and PSNFL figures are not zero). + - p.33 and p.109 (+0m in all six years; p.33's PSND and PSNFL figures aren't zero). - Table 6.1's footnote 2 sets the same bar for its lines: "AME and tax costings with an impact of less than +-£10 million in every year are not included". Yet two such lines appear: + In Table 6.1's footnote 2 HM Treasury sets the same bar for its lines: "AME and tax costings with an impact of less than +-£10 million in every year are not included". Yet it prints two such lines: - line 48 (0, then +5 a year; Spend head); - line 84 (0, -5, -5, -5, 0, 0; Spend head). - Lines 9 and 81 are also below £10m every year. The footnote's bar covers AME and tax costings only, and Table 6.1 prints no head finer than Tax or Spend to say whether these two are AME or departmental spending. -- **The lines 45 and 47 differences.** Table 6.1's footnote 4 points to this document for the packages' full costings, but the components do not sum to the lines (above), and no note explains the gap. - - For line 45, the gap has the sign the HMRC spending the notes describe would give: £70 million on debt collection (p.49) and £485 million on compliance staff (p.50), "over the next five years". The document does not say that this spending is the difference. + Lines 9 and 81 also fall below £10m every year. The footnote's bar covers AME and tax costings only, and HM Treasury prints no head in Table 6.1 finer than Tax or Spend to say whether these two count as AME or departmental spending. +- **The lines 45 and 47 differences.** In Table 6.1's footnote 4 HM Treasury points to this document for the packages' full costings, but the components don't sum to the lines (above), and it explains the gap in no note. + - For line 45, the gap has the sign that the HMRC spending HM Treasury describes in the notes would give: £70 million on debt collection (p.49) and £485 million on compliance staff (p.50), "over the next five years". HM Treasury doesn't say that this spending makes up the difference. - For line 47, the gap changes sign: +205 in 2026-27, -135 in 2027-28. -- **Notes with no Table 6.1 title.** BBB defence (p.41) and Erasmus+ (p.94) have costing notes, but no Table 6.1 title names them. +- **Notes with no Table 6.1 title.** HM Treasury prints costing notes for BBB defence (p.41) and Erasmus+ (p.94), but names them in no Table 6.1 title. - **Minor title and wording differences:** - - p.77's title says "from 25 June to 1 September", while Table 6.1 says "from 25 June 2026 to 1 September 2026". - - p.16's title says "Weekly Earning Limit", while its text says "Weekly Earnings Limit" (Table 6.1 follows the title). + - HM Treasury writes "from 25 June to 1 September" in p.77's title and "from 25 June 2026 to 1 September 2026" in Table 6.1. + - On p.16 HM Treasury writes "Weekly Earning Limit" in the title and "Weekly Earnings Limit" in the text (in Table 6.1 it follows the title). ## Not staged, and why -The family stages every Exchequer table and every printed estimate in the notes' text: tax bases, behavioural parameters, eligible counts, PSND, PSNFL and financial-transaction impacts. The appendix lists everything else page by page. +We stage every Exchequer table and every estimate HM Treasury prints in the notes' text: tax bases, behavioural parameters, eligible counts, PSND, PSNFL and financial-transaction impacts. We list everything else page by page in the appendix. -- **Design (README rule 8):** a measure's own rates, thresholds, payment amounts, dates and durations, and counts of products, sites or bodies. They stay in `reform_hint` and `note`. +- **Design (README rule 8):** a measure's own rates, thresholds, payment amounts, dates and durations, and counts of products, sites or bodies. We keep them in `reform_hint` and `note`. - **Contingent liabilities (rule 8):** - p.42's "£10 billion in new guarantees are issued each year between 2027-28 and 2030-31" (UK Export Finance); - p.83's guaranteed lending "by £1.0 billion in 2026-27, £1.5 billion in 2027-28 and £2.0 billion in each of 2028-29 and 2029-30" and its £6.5 billion total (Growth Guarantee Scheme). - Both are the volumes on which the notes cost fees and losses. Each note's `note` keeps the sentence. + In both, HM Treasury gives the volumes on which the notes cost fees and losses. We keep the sentence in each note's `note`. - **Spending allocations to programmes and financing remits (rule 8):** - - p.34's split of the British Business Bank recapitalisation (£1,405 million for commercial equity, £165 million of CDFI lending and £450 million of venture debt, within c. £2 billion), kept in `note`; + - p.34's split of the British Business Bank recapitalisation (£1,405 million for commercial equity, £165 million of CDFI lending and £450 million of venture debt, within c. £2 billion), which we keep in `note`; - p.33's £1 billion reprofiled; p.35's £2bn Energy Wealth Fund; p.41's £450 million; p.42's £50 billion of capacity; - p.109's £700m of PWLB debt repaid; - HMRC's £70 million (p.49) and £485 million (p.50) and the staff numbers on pp.50 and 56. -- **Numbers printed only in titles:** AB2025 kept these in `reform_hint`, and Table 6.1 carries the same titles. +- **Numbers printed only in titles:** in AB2025 we kept these in `reform_hint`, and HM Treasury prints the same titles in Table 6.1. - **Baseline parameters:** p.91's RPI outturn and counterfactual maximum rate. -- **p.58's "-£35m in 2029-30 in expenditure terms":** the same figure as that note's Exchequer table cell for 2029-30, so staging it would count it twice. -- **p.94's "£0.6bn was allocated for the cost of Erasmus+ association for 2027-28":** an allocation made at Spring Forecast 2026, an earlier event, with a different scope from this note's student-loan and delivery costs. -- **Annex A (pp.110-121):** indexation assumptions in the forecast baseline, not measure costings. AB2025 did not stage them either. +- **p.58's "-£35m in 2029-30 in expenditure terms":** HM Treasury repeats that note's Exchequer table cell for 2029-30, so staging it would count it twice. +- **p.94's "£0.6bn was allocated for the cost of Erasmus+ association for 2027-28":** the government made this allocation at Spring Forecast 2026, an earlier event, with a different scope from this note's student-loan and delivery costs. +- **Annex A (pp.110-121):** HM Treasury gives indexation assumptions in the forecast baseline, not measure costings. We didn't stage them in AB2025 either. - **The introduction (p.10):** the £10 million negligible threshold, and the list of five measures whose methodology "the government will publish further information on ... by the end of the year": - More Timely Payment, Distributions, Foreign Sovereign Investments, Founders' Relief and BADR. - - Those publications will be new primaries when they appear. + - We'll treat those publications as new primaries when they appear. - **References, not quantities:** scheme, form and statute numbers such as EMI2, Plan 2 and Plan 3, form P87, Class 1 and 1A, Group 7 of Schedule 9 to the VAT Act 1994, FRS 102 and the Five-Year Strategic Plan. -- **Uncertainty ratings:** the document prints none. Each row's `note` keeps the note's "Areas of uncertainty" text. p.101 says "there is no uncertainty over the figures involved". The OBR's ratings belong to uk_obr_tables. -- **Charts:** the document has none. +- **Uncertainty ratings:** HM Treasury prints none, so we keep each note's "Areas of uncertainty" text in the row's `note`. On p.101 HM Treasury says "there is no uncertainty over the figures involved". The OBR's ratings belong in uk_obr_tables. +- **Charts:** HM Treasury prints none. - **Unreached pages:** none. ### Appendix: every unstaged number, by page @@ -355,7 +362,7 @@ D marks design, C a contingent liability, S a spending allocation, input or meas - £700m of PWLB debt repaid, £500m for Woking and £200m for Thurrock (S); - two local authorities; a transfer between two central government bodies (D). -## Proposals used +## Proposals we use - `proposed_metric`: - Reused from earlier families: `behavioural_parameter` (8 rows), `tax_gap` (2, the uk_rf_ab2025 spelling), `tax_base` (1) and `financial_transaction_cash_impact` (1, the uk_hmt_tables spelling). @@ -364,17 +371,17 @@ D marks design, C a contingent liability, S a spending allocation, input or meas - `psnfl_impact` (2): its impact on net financial liabilities; - `psnfl_fair_value_impact` (2): the fair-value part of the PSNFL impact beyond PSND. - They parallel the existing `psnd_share_of_gdp` and `psnfl_share_of_gdp`. + We define them as we define the existing `psnd_share_of_gdp` and `psnfl_share_of_gdp`. - `proposed_unit`: `elasticity` (1). -- Registered values used beyond `revenue_change` and `gbp`: +- Registered values we use beyond `revenue_change` and `gbp`: - `eligible_count` with `persons`; - `percent`; - - `properties`, a unit registered since the AB2025 harvest proposed it. + - `properties`, a unit we have listed in the registry since we proposed it in the AB2025 harvest. - `proposed_baseline`: none. ## Attribution -- **`own` and `different_model`:** 595 rows. They are HMT and HMRC costings that the OBR certified ("All of these costings were submitted to the independent Office for Budget Responsibility (OBR) for their certification", p.10). +- **`own` and `different_model`:** 595 rows. HMT and HMRC made these costings, and the OBR certified them ("All of these costings were submitted to the independent Office for Budget Responsibility (OBR) for their certification", p.10). - **`restated` and `administrative_fact`:** the two tax gap rows. ## Distinct condition values @@ -391,29 +398,29 @@ D marks design, C a contingent liability, S a spending allocation, input or meas - not stated (the note's costing section does not use the word 'behaviour' or 'behavioural'); - 13 distinct "none: ''". - `horizon`: the period to 2031-32 (cumulative) -- Also used: `table_61_line`, `proposed_measure_key`, `document_position`, `fiscally_neutral`, `static_impact`, `table_footnote`, `statistic` (12 labels) and `parameter` (5 labels). +- We also use: `table_61_line`, `document_position`, `fiscally_neutral`, `static_impact`, `table_footnote`, `statistic` (12 labels) and `parameter` (5 labels). ## Review fixes, 3 November 2026 -The costings review (`harvest-staging/ab2026/reviews/verify-costings.md` on the staging branch) approved this family with fixes. They sit in an explicit block before `build()` in `stage.py`. No primary was re-opened (the session's network policy refused gov.uk); no printed value changed. +In its review (`harvest-staging/ab2026/reviews/verify-costings.md` on the staging branch), the costings reviewer approved this family with fixes. We put them in an explicit block before `build()` in `stage.py`. We re-opened no primary, because the session's network policy refused gov.uk, and changed no printed value. -- **Finding 1:** p.50's UK tax gap rows (6.4% and £59.2bn) are context: `measure_key` and `reform_hint` null, no `row_type`, `component` or `table_61_line`. They stay `restated`, `administrative_fact`, `administrative_data`, fy 2024-25. Registry keys now cover 573 rows, and components 157. -- **Finding 2:** p.41 keeps its proposed key; the double-counting caveat is in its `mapping_note`, the p.41 bullet and the proposed-key table. -- **Finding 3:** the line 45 and 47 component mapping notes now say their notes do not sum to the line, with Table 6.1's differences. -- **Finding 4:** no action. The coordinator found that the lane's copies of `pipeline/ab2026_harvest_contract.py` and the harvest README match #231's head byte for byte, so only this family directory moved. -- **Finding 5:** no change (p.84's £20m and p.58's +£130m are accepted as staged). -- **Verification:** the contract checker reports 597 rows, 0 problems, and a re-run of `stage.py` reproduces the gzip byte for byte. COUNTS.json pins its sha256. +- Finding 1: we made p.50's UK tax gap rows (6.4% and £59.2bn) context, with `measure_key` and `reform_hint` null and no `row_type`, `component` or `table_61_line`. We keep them `restated`, `administrative_fact`, `administrative_data`, fy 2024-25. Registry keys then covered 573 rows, and components 157. +- Finding 2: we kept p.41's proposed key, and stated the double-counting caveat in its `mapping_note`, the p.41 bullet and the table of keys with no Table 6.1 line. +- Finding 3: in the line 45 and 47 component mapping notes we now say that their notes don't sum to the line, and give Table 6.1's differences. +- Finding 4: we took no action. The coordinator found that the lane's copies of `pipeline/ab2026_harvest_contract.py` and the harvest README match #231's head byte for byte, so we moved only this family directory. +- Finding 5: we made no change; the reviewer accepted p.84's £20m and p.58's +£130m as staged. +- The contract checker reports 597 rows, 0 problems, and when we re-run `stage.py` it reproduces the gzip byte for byte. COUNTS.json pins its sha256. ## Review fixes, 4 November 2026 -Review A (PR #231, issue comment 6635870934) approved this family with fixes C-1 to C-7. They sit in an explicit block before `build()` in `stage.py`. No primary was re-opened; no printed value changed. - -- **C-1:** p.108's six Exchequer rows (the landlord fee's receipts, 585–590) are `row_type: measure_component` with `component: "Receipts (fee income; offset by increased DEL spending, fiscally neutral)"`. Value, metric and proposed key stay. Components: 157 → 163. -- **C-2:** p.50's two context rows (233–234) no longer say they are "a tax-base input of the measure"; their note says they are context figures with null key and reform_hint (AB2025 rule 3). -- **C-3:** p.41's three text rows (176–178) carry finding 2's double-counting sentence. -- **C-4:** the family follows rules 1 to 10; the Conventions bullet on rule 10 says why pp.34, 35 and 41's PSNB wording keeps `revenue_change`. -- **C-5:** the key tally names the 2 context rows. -- **C-6:** p.84's precedent is the AB2025 uk_obr_tables family, in NOTES and the row note. -- **C-7:** `stage.py` is on the staging branch. -- **T-3 (uk_hmt_tables):** `uk_hmt_tables` no longer emits `conditions.fiscal_head`, read from the workbook's hidden check column. The sentences here that reasoned from it (the p.41 bullet, lines 9, 48, 81 and 84, and the lines without a costing note) now rely only on Table 6.1's printed Tax and Spend heads. -- **D2-1 (GPT-6.1 Sol's delta review of these fixes):** C-1's step had marked p.108's whole costing record, which also feeds the note's text row, so row 584 (the 4.9 million properties in scope, `tax_base`) had become a receipts component. `stage.py` now marks only the six Exchequer rows, and row 584 matches its 4f424b9 version again. +Review A (PR #231, issue comment 6635870934) approved this family with fixes C-1 to C-7. We put them in an explicit block before `build()` in `stage.py`, re-opened no primary and changed no printed value. + +- C-1: we mark p.108's six Exchequer rows (the landlord fee's receipts, 585–590) `row_type: measure_component` with `component: "Receipts (fee income; offset by increased DEL spending, fiscally neutral)"`. We kept the value, metric and proposed key. Components: 157 → 163. +- C-2: we no longer call p.50's two context rows (233–234) "a tax-base input of the measure"; in their note we now call them context figures with null key and reform_hint (AB2025 rule 3). +- C-3: we added finding 2's double-counting sentence to p.41's three text rows (176–178). +- C-4: we follow rules 1 to 10, and in the Conventions bullet on rule 10 we say why pp.34, 35 and 41's PSNB wording keeps `revenue_change`. +- C-5: we name the 2 context rows in the key tally. +- C-6: we name the AB2025 uk_obr_tables family as p.84's precedent, in NOTES and the row note. +- C-7: we keep `stage.py` on the staging branch. +- T-3 (uk_hmt_tables): `uk_hmt_tables` no longer emits `conditions.fiscal_head`, which it read from the workbook's hidden check column. In the sentences here that reasoned from it (the p.41 bullet, lines 9, 48, 81 and 84, and the lines without a costing note), we now rely only on Table 6.1's printed Tax and Spend heads. +- D2-1 (GPT-6.1 Sol's delta review of these fixes): C-1's step had marked p.108's whole costing record, from which we also build the note's text row, so row 584 (the 4.9 million properties in scope, `tax_base`) had become a receipts component. `stage.py` now marks only the six Exchequer rows, and row 584 matches its 4f424b9 version again. diff --git a/sources/harvest-uk-ab2026-2026-10-29/uk_hmt_redbook/NOTES.md b/sources/harvest-uk-ab2026-2026-10-29/uk_hmt_redbook/NOTES.md index d471a0dd4..5277c662a 100644 --- a/sources/harvest-uk-ab2026-2026-10-29/uk_hmt_redbook/NOTES.md +++ b/sources/harvest-uk-ab2026-2026-10-29/uk_hmt_redbook/NOTES.md @@ -1,10 +1,10 @@ # uk_hmt_redbook: Autumn Budget 2026 red book, Impact on households and Pension Tax Allowance explainer -**Working files.** `stage.py`, which writes this family's two data files and is the file meant wherever these notes say `stage.py`, lives on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/hmt-redbook/uk_hmt_redbook/stage.py`; this branch carries only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz`. `.ab2026-primaries/uk_hmt_redbook/` was the lane's local primaries folder and was never committed: the repository carries no primary bytes, so fetch each primary from its `manifest.jsonl` URL and check its sha256. +We keep `stage.py`, which writes this family's two data files, on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/hmt-redbook/uk_hmt_redbook/stage.py`. We commit only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz` to this branch, and no primary bytes to the repository. The lane kept its local primaries in `.ab2026-primaries/uk_hmt_redbook/` and never committed them, so fetch each primary from its `manifest.jsonl` URL and check its sha256. -Source `hm_treasury`. **425 rows from 3 cited primaries (6 hashed versions), 21 proposed measure keys.** `stage.py` embeds every row and manifest line as data; `python3 stage.py` rewrites `claims_staged.jsonl.gz` and `manifest.jsonl` byte for byte with the standard library alone. No primary bytes are vendored. +We use source `hm_treasury`. We stage **425 rows from 3 cited primaries (6 hashed versions), on 50 measure keys, 21 of which we proposed and registered in tranche 2.** We embedded every row and manifest line in `stage.py` as data. `python3 stage.py` rewrites `claims_staged.jsonl.gz` and `manifest.jsonl` byte for byte with the standard library alone. We vendor no primary bytes. -The first staging (29 October, morning) produced 1,699 rows and 250 proposed keys. This rework applies README rules 8 and 9: it keeps estimates of effects, drops design (parameters, allocations, financing, contingent liabilities), and gives each printed part of one Table 6.1 measure that measure's key. Review A at 4f424b9 (4 November) dropped 7 more (see "Review A fixes"). The section "Not staged (rules 8 and 9)" accounts for all 1,274 dropped rows. +Our first staging (29 October, morning) produced 1,699 rows and 250 proposed keys. In this rework we apply README rules 8 and 9. We keep estimates of effects and drop design (parameters, allocations, financing, contingent liabilities), and we key each printed part of one Table 6.1 measure to that measure. After review A at 4f424b9 (4 November) we dropped 7 more (see "Review A fixes"). We account for all 1,274 dropped rows in the section "Not staged (rules 8 and 9)". ## Primaries @@ -19,29 +19,29 @@ The first staging (29 October, morning) produced 1,699 rows and 250 proposed key ## Access recipe -Every fresh fetch used: +For every fresh fetch the lane ran: ```sh curl -sSL -A "Mozilla/5.0 (Macintosh; Intel Mac OS X 14_0) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/17.0 Safari/605.1.15" URL -o PATH shasum -a 256 PATH ``` -All fresh fetches returned HTTP 200, and the files sit under `.ab2026-primaries/uk_hmt_redbook/`. No page refused access and none went unreached. The manifest labels the two historical copies `local copy`; they are kept for the changed-primary comparison (README rule 7), not as blocked-fetch fallbacks. +Every fresh fetch returned HTTP 200, the lane saved the files under `.ab2026-primaries/uk_hmt_redbook/`, and no page refused access. We label the two historical copies `local copy` in the manifest. We keep them for the changed-primary comparison (README rule 7), not as blocked-fetch fallbacks. -- **Red book:** fresh PDF fetched about 2026-10-29T15:41Z. `19b337ac…` matches the supplied 29 October download. `pdftotext -layout` and `diff -u` against the supplied 28 October version `bed70546…` find one change: paragraph 6.207's heading lost the words "and potential for ULEV SalSac reforms", with resulting line wrapping. **No printed number changed.** Every red-book row cites the current version. Printed page = PDF page minus six. -- **HTML discovery:** the publication content API `https://www.gov.uk/api/content/government/publications/budget-2026-document` (hash `4568690c1ca115ff4a7aa5aa116bcdb39cebee39d8af4490b8dc3ad544b3ec6d`) names the HTML attachment, fetched as a cross-check (hash `33ee5a00b6383104f83790719c0a67e396cfb52e7ea93986b76ab15a19fdaaec`). Neither supplies rows; the PDF is the primary. -- **Impact on households:** fresh PDF fetched 2026-10-29T15:41:44Z, identical to the supplied 29 October copy, read with `pdftotext -layout`. Printed pages equal PDF pages. -- **Pension Tax Allowance explainer:** fresh content API JSON fetched 2026-10-29T15:41:44Z, matching the supplied 28 October response. Rows quote its decoded `details.body`. **The supplied 29 October response differs:** its body is a 353-character introduction without the counts, worked cases or FAQ of the 19,727-character full body. Both carry publication date 28 October. No row cites the abbreviated version. -- **Pension Tax Allowance explainer, republished (README rule 7).** Review A fetched the content API on 4 November and got `83f8c894…` (24,356 bytes, `updated_at` 2026-10-29T15:55:28Z, a 19,206-character body): HM Treasury republished the explainer about 14 minutes after the lane's fetch. All twelve explainer rows' quotes and values are verbatim in it. It fixes "part-time wok" and no longer repeats an FAQ question; "£13.319" remains. The rows keep citing `cc3ec22e…`, the version they were read from. +- **Red book.** The lane fetched a fresh PDF about 2026-10-29T15:41Z, and `19b337ac…` matches the supplied 29 October download. It ran `pdftotext -layout` and `diff -u` against the supplied 28 October version `bed70546…` and found one change: HM Treasury cut the words "and potential for ULEV SalSac reforms" from paragraph 6.207's heading, which rewrapped the lines. **HM Treasury changed no printed number.** In every red-book row we cite the current version. The printed page equals the PDF page minus six. +- **HTML discovery.** The publication content API `https://www.gov.uk/api/content/government/publications/budget-2026-document` (hash `4568690c1ca115ff4a7aa5aa116bcdb39cebee39d8af4490b8dc3ad544b3ec6d`) names the HTML attachment, which we fetched as a cross-check (hash `33ee5a00b6383104f83790719c0a67e396cfb52e7ea93986b76ab15a19fdaaec`). We staged no rows from either and take the PDF as the primary. +- **Impact on households.** The lane fetched a fresh PDF at 2026-10-29T15:41:44Z, identical to the supplied 29 October copy, and read it with `pdftotext -layout`. Its printed pages equal PDF pages. +- **Pension Tax Allowance explainer.** The lane fetched fresh content API JSON at 2026-10-29T15:41:44Z, matching the supplied 28 October response, and in our rows we quote its decoded `details.body`. **The supplied 29 October response differs.** In it HM Treasury gives a 353-character introduction without the counts, worked cases or FAQ of the 19,727-character full body. HM Treasury gives publication date 28 October in both. We cite the abbreviated version in no row. +- **Pension Tax Allowance explainer, republished (README rule 7).** Review A fetched the content API on 4 November and got `83f8c894…` (24,356 bytes, `updated_at` 2026-10-29T15:55:28Z, a 19,206-character body). So HM Treasury republished the explainer about 14 minutes after the lane's fetch. All twelve explainer rows' quotes and values appear verbatim in it. HM Treasury fixed "part-time wok" and no longer repeats an FAQ question, but left "£13.319". We keep the rows citing `cc3ec22e…`, the version we read them from. ## Coverage -Rows run in document order: red book (printed page, then position), Impact on households, Pension Tax Allowance explainer. +We order rows by document: red book (printed page, then position), Impact on households, Pension Tax Allowance explainer. - **Red book (357 rows):** - - 160 fiscal aggregates restated from the OBR, ONS and IMF, with two shares of HM Treasury's own arithmetic (Box 1.C footnotes 54 and 56): Chapter 1 text, Chart 1.6's six data labels, Tables 1.2 and 1.3; - - 84 cells of Table 6.3, which breaks Table 6.1 line 45 into eleven items, with its subtotal, the line 46 memo and the subtotal including memo; - - 28 costs, yields and savings: Table 6.1 measures where the red book prints the line's own Exchequer impact, printed parts of those lines, and saving estimates with their own scopes; + - 160 fiscal aggregates that HM Treasury restates from the OBR, ONS and IMF, with two shares of HM Treasury's own arithmetic (Box 1.C footnotes 54 and 56): Chapter 1 text, Chart 1.6's six data labels, Tables 1.2 and 1.3; + - 84 cells of Table 6.3, in which HM Treasury breaks Table 6.1 line 45 into eleven items, with its subtotal, the line 46 memo and the subtotal including memo; + - 28 costs, yields and savings: Table 6.1 measures where HM Treasury prints the line's own Exchequer impact, printed parts of those lines, and saving estimates with their own scopes; - 85 effects on people and businesses: counts affected; gains, losses and bill changes; GDP, earnings, price and jobs effects; Barnett consequentials. - **Impact on households (56 rows):** the distributional and context statements of paragraphs 1.7, 1.9, 1.10 (first bullet), 2.4, 2.5 and 2.10, and the 50 cells of Table 2.C. - **Pension Tax Allowance explainer (12 rows):** three pensioner counts and nine worked-case outcomes. @@ -50,8 +50,8 @@ Rows by measure key: | measure_key | rows | of which components (rule 9) | |---|---:|---:| -| `null` with a proposed key | 66 | 18 | -| `null`, context (no proposed key) | 189 | 0 | +| the 21 keys we proposed (below) | 66 | 18 | +| `null`, context (no key) | 189 | 0 | | `ab2026__hmrc_tax_gap_package` | 73 | 67 | | `ab2026__pension_tax_allowance` | 12 | 0 | | `ab2026__british_industrial_competitiveness_scheme` | 8 | 0 | @@ -82,7 +82,7 @@ Rows by measure key: | `ab2026__national_wealth_fund_capital` | 1 | 0 | | `ab2026__parental_allowance_fathers_partners_adopters` | 1 | 0 | -The 189 context rows carry no proposed key (README rule 2): 135 fiscal aggregates about no measure (133 restated, and HM Treasury's own two Box 1.C shares), and 54 Impact on households rows (Table 2.C and the context statements of paragraphs 1.10, 2.4, 2.5 and 2.10). The other 24 Table 1.3 rows, the OBR's effect of policy decisions, and Chart 1.6's "Effect of policy" carry `ab2026__package_government_decisions_since_march_2026`. +We give the 189 context rows no key (README rule 2). They comprise 135 fiscal aggregates about no measure (133 restated, plus HM Treasury's own two Box 1.C shares) and 54 Impact on households rows (Table 2.C and the context statements of paragraphs 1.10, 2.4, 2.5 and 2.10). We key the other 24 Table 1.3 rows, the OBR's effect of policy decisions, and Chart 1.6's "Effect of policy" to `ab2026__package_government_decisions_since_march_2026`. Rows by metric: @@ -125,36 +125,47 @@ Rows by metric: | proposed `share_in_top_half_of_expenditure` | 1 | | proposed `share_with_zero_liability` | 1 | -Attribution: 257 rows `own`, 168 `restated` (OBR, ONS, IMF, DWP, DBT, the British Business Bank, the National Wealth Fund). +We mark 257 rows `own` and 168 `restated` (OBR, ONS, IMF, DWP, DBT, the British Business Bank, the National Wealth Fund). ## Conventions -- **Values as printed.** `value_raw` is the printed string, and `value` scales it to the unit (£ million × 1e6, £ billion × 1e9; percent stays percent; words such as "three-quarters" become fractions with unit `share`). Bounds keep the printed qualifier: "over" and "more than" are `range_low`, "up to" is `range_high`, "around" is `point`. -- **Sign.** Values are never re-signed. `conditions.sign_convention` states each row's direction in the phrasing `scorecard_db/signs.py` reads: +- **Values as printed.** We copy the printed string to `value_raw` and scale it to the unit in `value` (£ million × 1e6, £ billion × 1e9; percent stays percent; words such as "three-quarters" become fractions with unit `share`). We keep the printed qualifier on bounds: we map "over" and "more than" to `range_low`, "up to" to `range_high`, and "around" to `point`. +- **Sign.** We never re-sign values. In `conditions.sign_convention` we state each row's direction in the phrasing `scorecard_db/signs.py` reads: - Exchequer costs: "positive = cost to the Exchequer (as printed; not re-signed)"; - savings: "positive = saving to the Exchequer"; - yields, including Table 6.3: "positive = yield to the Exchequer (reduces borrowing)"; - - Table 1.3: "positive = increase in PSNB" (the opposite of Table 6.1's yield convention), except on its two memo lines, which give the change in the current budget deficit: "positive = increase in current budget deficit" (README rule 10, since 3 November); + - Table 1.3: "positive = increase in PSNB", the opposite of Table 6.1's yield convention. On its two memo lines HM Treasury gives the change in the current budget deficit: "positive = increase in current budget deficit" (README rule 10, since 3 November); - Table 1.2: deficits and debt positive, surpluses negative, as printed; - - Chart 1.6's three bridge steps (rows 21, 22 and 24 at 4f424b9): "positive = reduction in current budget deficit (increase in stability rule headroom)", the reverse of Table 1.3's memo, as printed (Chart 1.6's "Effect of policy" +2.0 is Table 1.3's policy memo −2.0 in 2029-30); - - Box 1.C: each row says what it measures and which way it runs ("positive = fall in borrowing (… as worded: fell by)", "as printed; positive = current budget deficit (% of GDP)" and so on; review A's RB-12); + - Chart 1.6's three bridge steps (rows 21, 22 and 24 at 4f424b9): "positive = reduction in current budget deficit (increase in stability rule headroom)". HM Treasury prints these steps with the opposite sign to Table 1.3's memo, for example +2.0 for Chart 1.6's "Effect of policy" and −2.0 for the policy memo in 2029-30; + - Box 1.C: we state each figure's meaning and direction on its row (review A's RB-12), for example "positive = fall in borrowing (… as worded: fell by)" and "as printed; positive = current budget deficit (% of GDP)"; - Barnett consequentials: "positive = increase in devolved government funding"; - gains to a person or business: "positive = gain to the (…, as printed)", and row 185's CPI effect "positive = reduction in CPI inflation (as worded: reduce inflation by)" (RB-17). -- **Periods.** Fiscal-year figures carry `conditions.fy`, end-year `period` and `time_basis: fiscal_year`. A multi-year total, or an average over the forecast, keeps its whole horizon in `conditions.horizon`, sits at its final fiscal year and is never annualised; "over the forecast" means 2026-27 to 2031-32, as paragraphs 1.20 and 1.40 date it. A figure printed more than once keeps a row for each printing (AB2025 did the same), and the repeats share one dating, the most specific mention's (review A's RB-15); where more than two printings would share one claim identity, only two are kept (RB-2). An undated figure ("by the late 2030s", "this winter") sits at the publication year 2026 with the printed horizon in conditions and, where the year matters, `parse_confidence: medium`. -- **Geography** comes from the harvest's closed list (`UK`, `GB`, `England`, `Wales`, `Scotland`, `Northern Ireland`, `England and Wales`, `Scotland, Wales and Northern Ireland`); any detail goes in `conditions.geography_note` or `note`. The Ofgem price cap and BICS are `GB`. -- **Prices.** Every money figure is cash, as printed; Table 1.3's rows carry `price_basis: current prices`. No basis is converted. (The first staging tagged the £595 million administration saving `real terms`; paragraph 2.14's "real terms" belongs to the Spending Review's 16% cut, and the £595 million equals Table 6.1 line 63's cash +595.) -- **"Central".** HM Treasury prints single estimates; `point` means the printed figure. No row is an inferred scenario. The one OBR scenario row (borrowing £12 billion lower if energy prices follow a lower path) carries `conditions.scenario`. -- **Rule 9 components.** A printed part of one Table 6.1 measure carries the line's key, `conditions.component` naming the part and `conditions.row_type: measure_component`. Components are never summed with the line's total. This family has 102: the 66 Table 6.3 item cells and the 160,000 employees (line 45); the £3 billion and £1.7 billion parts of line 41's £4.7 billion; the four food-tariff consumer savings (line 10; the June-and-November range contains the November-only one, so never add the two); the India agreement's UK, Scotland, Wales and Northern Ireland GDP effects (line 77; the UK figure is the trade deal alone, without the Double Contributions Convention); paragraph 3.22's six figures for line 14's real-terms freeze and SSB extension; the British Business Bank's 40,000 jobs (line 29's capacity half, as Policy Costings p.34); and, under proposed keys, the Tax and Spending parts of the OBR's policy effect, the three devolved Barnett consequentials and three parts of the regulatory administrative savings. -- **`component` means a part only.** Where the first staging used `component` for a Chapter 6 paragraph title on a whole-measure row, the condition is now `paragraph_heading`; in the pension worked cases it is `quantity`. Chart 1.6's six headroom-bridge data labels name their bar in `chart_bar` (changed on 3 November; the rework had left them under `component`). -- **Attribution.** OBR forecasts are `restated` with source model `obr_efo_forecast`; ONS, IMF, DWP, DBT and British Business Bank figures are `restated` and their notes name the original. HM Treasury's own costings, counts and worked cases are `own`. +- **Periods.** + - We give fiscal-year figures `conditions.fy`, end-year `period` and `time_basis: fiscal_year`. + - For a multi-year total, or an average over the forecast, we keep the whole horizon in `conditions.horizon`, put the row at its final fiscal year and never annualise it. We read "over the forecast" as 2026-27 to 2031-32, as HM Treasury dates it in paragraphs 1.20 and 1.40. + - Where HM Treasury prints a figure more than once, we keep a row for each printing (as in AB2025) and give the repeats one dating, the most specific mention's (review A's RB-15). Where more than two printings would share one claim identity, we keep only two (RB-2). + - We put an undated figure ("by the late 2030s", "this winter") at the publication year 2026, with the printed horizon in conditions and, where the year matters, `parse_confidence: medium`. +- **Geography.** We take it from the harvest's closed list (`UK`, `GB`, `England`, `Wales`, `Scotland`, `Northern Ireland`, `England and Wales`, `Scotland, Wales and Northern Ireland`), and put any detail in `conditions.geography_note` or `note`. We record the Ofgem price cap and BICS as `GB`. +- **Prices.** HM Treasury prints every money figure in cash, and we stage it so; on Table 1.3's rows we set `price_basis: current prices`. We convert no basis. (Our first staging tagged the £595 million administration saving `real terms`. But in paragraph 2.14 HM Treasury applies "real terms" to the Spending Review's 16% cut, and the £595 million equals Table 6.1 line 63's cash +595.) +- **"Central".** HM Treasury prints single estimates, and we mark the printed figure `point`. We infer no scenario on any row. On the one OBR scenario row (borrowing £12 billion lower if energy prices follow a lower path) we set `conditions.scenario`. +- **Rule 9 components.** We key a printed part of one Table 6.1 measure to the line, name the part in `conditions.component` and set `conditions.row_type: measure_component`. We never sum components with the line's total. In this family we stage 102: + - the 66 Table 6.3 item cells and the 160,000 employees (line 45); + - the £3 billion and £1.7 billion parts of line 41's £4.7 billion; + - the four food-tariff consumer savings (line 10). The June-and-November range contains the November-only one, so never add the two; + - the India agreement's UK, Scotland, Wales and Northern Ireland GDP effects (line 77). HM Treasury's UK figure covers the trade deal alone, without the Double Contributions Convention; + - paragraph 3.22's six figures for line 14's real-terms freeze and SSB extension; + - the British Business Bank's 40,000 jobs (line 29's capacity half, as in Policy Costings p.34); + - under the keys we proposed, the Tax and Spending parts of the OBR's policy effect, the three devolved Barnett consequentials and three parts of the regulatory administrative savings. +- **We use `component` for a part only.** Where our first staging used `component` for a Chapter 6 paragraph title on a whole-measure row, we now use `paragraph_heading`, and in the pension worked cases `quantity`. On 3 November we moved Chart 1.6's six headroom-bridge data labels, which the rework had left under `component`, to `chart_bar`. +- **Attribution.** We mark OBR forecasts `restated` with source model `obr_efo_forecast`. We mark ONS, IMF, DWP, DBT and British Business Bank figures `restated` and name the original in their notes. We mark HM Treasury's own costings, counts and worked cases `own`. ### Which spending figures stay (rule 8 exception) -A spending figure is a measure's design unless the red book presents it as the measure's Exchequer impact. The test applied: the figure stays when it equals the printed Table 6.1 line (from `uk_hmt_tables`), within the red book's rounding, as a total over the years the red book states or, for a per-year amount, as the line's per-year amount (review A's RB-20(b): the early-years books rows pass only the second form). Tax costs and yields, savings estimates and Table 6.3 stay regardless. +We treat a spending figure as a measure's design unless HM Treasury presents it as the measure's Exchequer impact. We keep the figure when it equals the printed Table 6.1 line (from `uk_hmt_tables`), within the red book's rounding. It may match as a total over the years HM Treasury states or, for a per-year amount, as the line's per-year amount (review A's RB-20(b): the early-years books rows pass only the second form). We keep tax costs and yields, savings estimates and Table 6.3 regardless. -| Table 6.1 line | Red book | Table 6.1 prints (£ million) | Rows | +| Table 6.1 line | Red book | Table 6.1 (£ million) | Rows | |---|---|---|---| -| 1 Household Energy Payment | £2.6 billion | -2,625 in 2026-27 | kept (2): the executive summary and para 2.48 (review A's RB-2 dropped Box 3.A's and para 3.5's repeats) | +| 1 Household Energy Payment | £2.6 billion | -2,625 in 2026-27 | kept (2): the executive summary and para 2.48 (under review A's RB-2 we dropped Box 3.A's and para 3.5's repeats) | | 9 Early years books | £5 million a year | -5 a year (2026-27 to 2028-29) | kept (2): the per-year amount equals the line's; the years differ, see anomalies | | 19 Small Producer Relief | £10 million per year | -10 a year from 2027-28 | kept (2) | | 32 EV charging | £100 million over two years | -50, -50 | kept (1) | @@ -175,99 +186,139 @@ A spending figure is a measure's design unless the red book presents it as the m | 37 Your First Home (Table 6.2 line 5) | £4.7 billion | -4,700 of financial transactions | dropped: a financial-transactions allocation | | 6.2 line 4 Sovereign AI | £60 million from 2027-28 | -45 a year | dropped | -## Proposed measure keys (21) +## Keys we proposed (21) -Each key below is the whole scope of its rows, stated as Table 6.1 lines or "no Table 6.1 line", with the red book's words. +We proposed each key below, and in tranche 2 we registered it and moved it into `measure_key`. We give each key the whole scope of its rows, as Table 6.1 lines or "no Table 6.1 line", with the red book's words. -- `ab2026__package_government_decisions_since_march_2026` (25 rows): the OBR's "Total effect of policy decisions" on PSNB in Table 1.3, with its Tax and Spending parts (components) and the memo effect on the current budget deficit, 2026-27 to 2031-32, and Chart 1.6's "Effect of policy" on stability rule headroom in 2029-30 (+2.0: the memo's −2.0 seen as headroom; proposed `fiscal_balance_change`). Crosses Table 6.1 lines 1–98 and adds the OBR's indirect effects; Table 1.3 footnote 1 says the OBR's EFO Table 3.1 explains the "Differences between the direct impacts of the policy measures as set out in Table 6.1 of the Budget, and how the OBR incorporate these into their forecast". Not `ab2026__package_total_policy_decisions`. -- `ab2026__package_hmrc_tax_gap_and_online_marketplace` (7 rows): exactly Table 6.1 lines 45 and 46. Table 6.3's "Subtotal including memo" (six years) and footnote 73's "£2.3 billion from this Budget": Table 6.3 prints +2,250m for lines 45 and 46 in 2029-30, against +1,980m for line 45 alone. +- `ab2026__package_government_decisions_since_march_2026` (25 rows): the OBR's "Total effect of policy decisions" on PSNB in Table 1.3, with its Tax and Spending parts (components) and the memo effect on the current budget deficit, 2026-27 to 2031-32. It also covers Chart 1.6's "Effect of policy" on stability rule headroom in 2029-30 (+2.0: the memo's −2.0 seen as headroom; proposed `fiscal_balance_change`). We scope the key across Table 6.1 lines 1–98 and the OBR's indirect effects. In Table 1.3 footnote 1 HM Treasury says the OBR's EFO Table 3.1 explains the "Differences between the direct impacts of the policy measures as set out in Table 6.1 of the Budget, and how the OBR incorporate these into their forecast". We don't use `ab2026__package_total_policy_decisions` here. +- `ab2026__package_hmrc_tax_gap_and_online_marketplace` (7 rows): exactly Table 6.1 lines 45 and 46. It covers Table 6.3's "Subtotal including memo" (six years) and footnote 73's "£2.3 billion from this Budget". In Table 6.3 HM Treasury prints +2,250m for lines 45 and 46 in 2029-30, against +1,980m for line 45 alone. - `ab2026__package_parliament_tax_gap_measures` (1 row): lines 45 and 46 plus the tax-gap measures of Autumn Budget 2024, Spring Statement 2025 and Budget 2025 (footnote 73). Paragraph 2.46: "the total due to be raised from measures to close the tax gap announced during this Parliament to over £12 billion in 2029-30". - `ab2026__package_property_and_savings_income_rates` (2 rows): exactly Table 6.1 lines 61 and 62. Paragraph 2.50: "84% of income tax payers will be unaffected by these changes"; "Around half of the revenue raised from this change is expected to come from the top 10% of households". -- `ab2026__package_breathing_space_for_households` (1 row): paragraph 3.2's "Government action at this Budget to provide breathing space for households", which the OBR expects to reduce CPI inflation by 0.1 percentage points in 2027. Crosses Table 6.1 lines; the red book lists none for this figure. Three scopes are in play, and the row keeps this key until tranche 2 settles them (the coordinator's call on review A's RB-17): Box 3.A's "breathing space for households" is lines 1 and 2 (the Household Energy Payment and the VAT cut); the OBR family keys its own CPI effect to lines 2 and 3 (`ab2026__package_fuel_duty_and_electricity_vat`); and Table 6.1 groups lines 1–11 under "Breathing space for families". Paragraph 1.38 repeats the 0.1 percentage points and is not staged. +- `ab2026__package_breathing_space_for_households` (1 row): paragraph 3.2's "Government action at this Budget to provide breathing space for households", which the OBR expects to reduce CPI inflation by 0.1 percentage points in 2027. We scope the key across Table 6.1 lines, though HM Treasury lists none for this figure. + - The row fits three scopes, and we keep it on this key (the coordinator's call on review A's RB-17). In tranche 2 we register the key with a `partial` verdict and leave the three readings open (`KEY_DECISIONS.md`), and at ingest we drop the row as a restatement of the OBR's figure. + - In Box 3.A HM Treasury's "breathing space for households" covers lines 1 and 2 (the Household Energy Payment and the VAT cut). In the OBR family we key the OBR's own CPI effect to lines 2 and 3 (`ab2026__package_fuel_duty_and_electricity_vat`). In Table 6.1 HM Treasury groups lines 1–11 under "Breathing space for families". + - HM Treasury repeats the 0.1 percentage points in paragraph 1.38, which we don't stage. - `ab2026__energy_debt_circuit_breaker` (3 rows): no Table 6.1 line. Paragraph 3.8: "a new debt strategy to address this and act as a circuit breaker on energy debt … via reforms to prevent debt building up and a one-off targeted debt relief scheme for eligible households". -- `ab2026__package_parliament_fuel_duty_decisions` (3 rows): Table 6.1 line 3 plus the fuel duty decisions of earlier events in this Parliament. Paragraph 3.13: "the government's decisions on fuel duty across this Parliament … compared with plans under the previous government"; footnote 98's baseline is the 5p cut reversed in April 2025 and annual RPI uprating. +- `ab2026__package_parliament_fuel_duty_decisions` (3 rows): Table 6.1 line 3 plus the fuel duty decisions of earlier events in this Parliament. Paragraph 3.13: "the government's decisions on fuel duty across this Parliament … compared with plans under the previous government". In footnote 98 HM Treasury takes as baseline the 5p cut reversed in April 2025 and annual RPI uprating. - `ab2026__minimum_wage_uprating_april_2027` (2 rows): no Table 6.1 line (a regulation). Paragraph 3.18: "raising the National Living Wage and National Minimum Wages" from 1 April 2027. -- `ab2026__package_business_rates_rhl_sbrr_and_transitional_support` (2 rows): Table 6.1 lines 13 and 14, the three bullets of paragraph 3.22 (SBRR thresholds; the real-terms freeze; the SSB extension), to which paragraph 3.23's "As a result of this package" refers. The red book does not say whether paragraph 3.21's pubs relief (line 12) also counts. -- `ab2026__package_pubs_relief_35pc_total` (1 row): the whole 35% business rates relief for pubs, social clubs and live music venues in England: the 15% announced in January 2026 (paragraph 6.189; no AB2026 Table 6.1 line) plus Table 6.1 line 12's further 20ppt. Paragraph 3.21: "Taken together, this relief is a tax cut worth an average of over £3,000 per pub next year." Not line 12's key, since line 12 is only the 20ppt (rule 3; review A's RB-3). -- `ab2026__package_regulation_tariff_and_sandbox_business_savings` (1 row): the executive summary's "almost £800 million per year through further cuts to the administrative burdens of regulation, extending suspensions on tariffs, and introducing new cross-economy powers so firms can safely test new technologies faster". Crosses Table 6.1 line 16 (business tariff suspensions) with measures that have no Table 6.1 line. -- `ab2026__regulatory_administrative_savings` (5 rows): no Table 6.1 line. Chapter 6 decision "Regulatory administrative savings" (paragraph 6.36) and paragraph 4.52's "up to an estimated £640 million per annum of new gross administrative savings for business"; its £290 million, £116 million and £105 million parts are components. -- `ab2026__rolls_royce_gas_turbine_programme` (1 row): no Table 6.1 line named. Paragraph 4.80: "co-investment in a joint, 10-year, £2.5‑billion programme for next-generation gas turbine technologies for civil and defence aerospace"; paragraph 4.83 says it "has not been contracted". -- `ab2026__package_defence_investment_plan` (1 row): the Defence Investment Plan published in June 2026, which no single line covers; lines 40–42 carry its savings and funding. Paragraph 5.34: "The Defence Investment Plan will create approximately 60,000 new jobs by 2029‑30 compared with 2023‑24." The OBR family's row on this key (paragraph 3.3's £1.8 billion a year) is the plan's RDEL funding part, a component (review A's K-2). -- `ab2026__package_dwp_fraud_and_error_measures` (1 row): the "new measures" of paragraph 5.17, "expected to save an additional £1.4 billion over the forecast" (2026-27 to 2031-32). The red book doesn't list them. Paragraphs 6.19–6.22 are line 47's own package (its footnote 4; Policy Costings pp.54–58, which `uk_hmt_costings` keys as line 47 components), but the £1.4 billion reconciles with no printed basis for line 47: the line prints +600m over the forecast, +940m with Table 6.2 line 9's +340m, and the costings sum to +570m. So the figure stays off line 47's key (rule 3; review A's RB-8). -- `ab2026__package_parliament_welfare_fraud_and_error_savings` (1 row): this Budget's new fraud and error measures plus those already in the forecast. Paragraph 5.17: "bringing total savings from fraud and error to over £20 billion over the forecast". It doesn't say which earlier measures count. (Review A suggested dropping "parliament" from the one-row key's name; tranche 2 can rename it when it registers the key.) -- `ab2026__package_barnett_consequentials` (4 rows): the Barnett consequentials of "decisions confirmed at Budget", 2026-27 to 2029-30, excluding the technical reduction from lower employer pension contributions (paragraph 4.150). Crosses Table 6.1's England spending lines; no line of its own. The three-government £1.3 billion is `row_type: total`, and the Scottish, Welsh and Northern Ireland figures are components, one per government; `uk_devolved_govts` types the Welsh Government's own figures as Welsh Government components of this key. All four rows say "positive = increase in devolved government funding" (review A's K-7(b)). +- `ab2026__package_business_rates_rhl_sbrr_and_transitional_support` (2 rows): Table 6.1 lines 13 and 14, the three bullets of paragraph 3.22 (SBRR thresholds; the real-terms freeze; the SSB extension), to which paragraph 3.23's "As a result of this package" refers. In the red book HM Treasury doesn't say whether paragraph 3.21's pubs relief (line 12) also counts. +- `ab2026__package_pubs_relief_35pc_total` (1 row): the whole 35% business rates relief for pubs, social clubs and live music venues in England. It adds the 15% announced in January 2026 (paragraph 6.189; no AB2026 Table 6.1 line) to Table 6.1 line 12's further 20ppt. Paragraph 3.21: "Taken together, this relief is a tax cut worth an average of over £3,000 per pub next year." We don't use line 12's key, since line 12 covers only the 20ppt (rule 3; review A's RB-3). +- `ab2026__package_regulation_tariff_and_sandbox_business_savings` (1 row): the executive summary's "almost £800 million per year through further cuts to the administrative burdens of regulation, extending suspensions on tariffs, and introducing new cross-economy powers so firms can safely test new technologies faster". We scope the key across Table 6.1 line 16 (business tariff suspensions) and measures that have no Table 6.1 line. +- `ab2026__regulatory_administrative_savings` (5 rows): no Table 6.1 line. It covers the Chapter 6 decision "Regulatory administrative savings" (paragraph 6.36) and paragraph 4.52's "up to an estimated £640 million per annum of new gross administrative savings for business". We stage its £290 million, £116 million and £105 million parts as components. +- `ab2026__rolls_royce_gas_turbine_programme` (1 row): HM Treasury names no Table 6.1 line. Paragraph 4.80: "co-investment in a joint, 10-year, £2.5‑billion programme for next-generation gas turbine technologies for civil and defence aerospace"; in paragraph 4.83 HM Treasury says it "has not been contracted". +- `ab2026__package_defence_investment_plan` (1 row): the Defence Investment Plan published in June 2026, which no single line covers; HM Treasury puts its savings and funding in lines 40–42. Paragraph 5.34: "The Defence Investment Plan will create approximately 60,000 new jobs by 2029‑30 compared with 2023‑24." In the OBR family we stage the row on this key (paragraph 3.3's £1.8 billion a year) as the plan's RDEL funding part, a component (review A's K-2). +- `ab2026__package_dwp_fraud_and_error_measures` (1 row): the "new measures" of paragraph 5.17, "expected to save an additional £1.4 billion over the forecast" (2026-27 to 2031-32). The red book doesn't list them. + - HM Treasury makes paragraphs 6.19–6.22 line 47's own package (its footnote 4; Policy Costings pp.54–58, which we key as line 47 components in `uk_hmt_costings`). + - But the £1.4 billion reconciles with no printed basis for line 47. HM Treasury prints +600m on Table 6.1 line 47 over the forecast, and +940m with Table 6.2 line 9's +340m, and the costings sum to +570m. So we keep the figure off line 47's key (rule 3; review A's RB-8). +- `ab2026__package_welfare_fraud_and_error_savings_in_forecast` (1 row): this Budget's new fraud and error measures plus those already in the forecast. Paragraph 5.17: "bringing total savings from fraud and error to over £20 billion over the forecast". HM Treasury doesn't say which earlier measures count. We proposed the key as `ab2026__package_parliament_welfare_fraud_and_error_savings`; review A suggested dropping "parliament" from the name, and tranche 2 renamed it when it registered the key. +- `ab2026__package_barnett_consequentials` (4 rows): the Barnett consequentials of "decisions confirmed at Budget", 2026-27 to 2029-30, excluding the technical reduction from lower employer pension contributions (paragraph 4.150). + - We scope the key across Table 6.1's England spending lines, and HM Treasury prints no line of its own for it. + - We mark the three-government £1.3 billion `row_type: total`, and the Scottish, Welsh and Northern Ireland figures as components, one per government. In `uk_devolved_govts` we type the Welsh Government's own figures as Welsh Government components of this key. + - On all four rows we say "positive = increase in devolved government funding" (review A's K-7(b)). - `ab2026__mod_counter_fraud_strategy` (1 row): no Table 6.1 line. Paragraph 5.32: "a new Counter Fraud Strategy in November, with investment in fraud and error analytics and closer working with the Public Sector Fraud Authority set to deliver £250 million of benefits". -- `ab2026__state_pension_triple_lock_adjustment` (2 rows): no Table 6.1 line (its savings fall after the forecast). Paragraph 5.23: "adjusting the Triple Lock from April 2030 so that the State Pension rises each year by the highest of 2.5%, inflation, or the amount required to maintain its value relative to earnings as reached in April 2029"; DWP's estimate (footnote 61). -- `ab2026__package_hmt_distributional_budget_2026` (1 row): HM Treasury's own grouping, the Budget 2026 policies modelled in Impact on households (paragraph 1.4): Table 6.1 lines 2, 3, 4, 5, 8, 11, 15, 26, 39, 49, 52, 59, 60, 61, 62, 72 and 95, plus the private landlord registration fee, which has no line. "Decisions announced at Budget 2026 (including those pre-announced where funding is allocated at Budget 2026)". It excludes line 1, the Household Energy Payment. -- `ab2026__package_hmt_distributional_since_ab2024` (1 row): HM Treasury's own grouping, the modelled policies and public service settlements "announced from Autumn Budget 2024 onwards up to and including Budget 2026" (paragraph 1.8): the lines above plus earlier events' measures, which are not Table 6.1 lines. +- `ab2026__state_pension_triple_lock_adjustment` (2 rows): no Table 6.1 line (its savings fall after the forecast). Paragraph 5.23: "adjusting the Triple Lock from April 2030 so that the State Pension rises each year by the highest of 2.5%, inflation, or the amount required to maintain its value relative to earnings as reached in April 2029"; DWP made the estimate (footnote 61). +- `ab2026__package_hmt_distributional_budget_2026` (1 row): HM Treasury's own grouping, the Budget 2026 policies it models in Impact on households (paragraph 1.4). It covers Table 6.1 lines 2, 3, 4, 5, 8, 11, 15, 26, 39, 49, 52, 59, 60, 61, 62, 72 and 95, plus the private landlord registration fee, which has no line. "Decisions announced at Budget 2026 (including those pre-announced where funding is allocated at Budget 2026)". HM Treasury leaves out line 1, the Household Energy Payment. +- `ab2026__package_hmt_distributional_since_ab2024` (1 row): HM Treasury's own grouping, the modelled policies and public service settlements "announced from Autumn Budget 2024 onwards up to and including Budget 2026" (paragraph 1.8). It covers the lines above plus earlier events' measures, which Table 6.1 doesn't list. -The coordinator's shared spellings: no kept row has the scope of `ab2026__gas_standing_charge_moved_to_unit_rates`, `ab2026__pwlb_debt_relief_councils_severe_distress`, `ab2026__private_landlord_register_fee_65`, `ab2026__british_business_bank_defence_commitments_accelerated` or `ab2026__package_household_energy_payment_and_electricity_vat`. The dropped rows on those subjects were design: the £108 current gas standing charge and the end-2027 delivery date (parameters), the £500 million and £200 million of PWLB debt relief (allocations to two councils, which Table 6.1's footnote 2 excludes as adding nothing to borrowing), the £65 fee (a parameter), and the "up to £5 billion" NWF and BBB defence commitment (an investment total). No red-book figure covers exactly lines 1 and 2, or exactly lines 1, 2 and 3 (`ab2026__package_household_energy_vat_and_fuel_duty`). +The coordinator set shared spellings, and no kept row has the scope of `ab2026__gas_standing_charge_moved_to_unit_rates`, `ab2026__pwlb_debt_relief_councils_severe_distress`, `ab2026__private_landlord_register_fee_65`, `ab2026__british_business_bank_defence_commitments_accelerated` or `ab2026__package_household_energy_payment_and_electricity_vat`. We dropped the rows on those subjects as design: +- the £108 current gas standing charge and the end-2027 delivery date (parameters); +- the £500 million and £200 million of PWLB debt relief (allocations to two councils, which HM Treasury excludes in Table 6.1's footnote 2 as adding nothing to borrowing); +- the £65 fee (a parameter); +- the "up to £5 billion" NWF and BBB defence commitment (an investment total). + +No red-book figure covers exactly lines 1 and 2, or exactly lines 1, 2 and 3 (`ab2026__package_household_energy_vat_and_fuel_duty`). ## Quantity and baseline proposals - **Proposed metrics:** - fiscal aggregates (restated): `government_borrowing`, `government_borrowing_change`, `government_debt`, `government_net_worth`, `fiscal_balance`, `fiscal_balance_change` (also Chart 1.6's three bridge steps, which change the stability rule's headroom, i.e. the 2029-30 current budget), `debt_interest_change`, `fiscal_consolidation_share`; `block_grant_effect` (AB2025 list); - - household and business effects: `household_tax_change` (AB2025 list; worked cases), `average_cumulative_gain` (multi-year savings), `aggregate_household_saving` (AB2025 spelling; consumer savings), `bill_change_percent` (business electricity bills), `gross_earnings_change` (AB2025 spelling), `business_cost_saving` (administrative savings to business; replaces `operating_cost_change`, which the enum defines as an official operating-cost line), `housing_supply_change` (the 20,000 additional homes), `jobs_created_or_supported` (the NWF's 100,000, the BBB's 40,000, Rolls-Royce's 3,300 and the DIP's 60,000; `affected_count` counts individuals affected, so jobs are not it, review A's RB-9); + - household and business effects: + - `household_tax_change` (AB2025 list; worked cases); + - `average_cumulative_gain` (multi-year savings); + - `aggregate_household_saving` (AB2025 spelling; consumer savings); + - `bill_change_percent` (business electricity bills); + - `gross_earnings_change` (AB2025 spelling); + - `business_cost_saving` (administrative savings to business; replaces `operating_cost_change`, which we define in the enum as an official operating-cost line); + - `housing_supply_change` (the 20,000 additional homes); + - `jobs_created_or_supported` (the NWF's 100,000, the BBB's 40,000, Rolls-Royce's 3,300 and the DIP's 60,000; we keep `affected_count` for individuals affected, not jobs; review A's RB-9); - shares with exact meanings: `unaffected_share` (the share printed as unaffected), `share_with_zero_liability` (half of RHL properties paying no rates), `share_in_non_gaining_deciles`, `share_in_net_recipient_deciles`, `income_mobility_share`, `share_in_top_half_of_expenditure` (Impact on households context). - **Proposed units:** `businesses` (15 rows), `dwellings` (1). -- **Proposed baselines:** `hmt_no_budget_2026_modelled_policies` (paragraph 1.6, "a counterfactual in which the modelled policies were not introduced") and `hmt_no_policy_changes_from_ab2024_through_ab2026` (paragraph 1.8, "a counterfactual in which the modelled policies and spending settlements were not introduced"), one row each. Other rows have `baseline_policy: null`. +- **Proposed baselines:** `hmt_no_budget_2026_modelled_policies` (paragraph 1.6, "a counterfactual in which the modelled policies were not introduced") and `hmt_no_policy_changes_from_ab2024_through_ab2026` (paragraph 1.8, "a counterfactual in which the modelled policies and spending settlements were not introduced"), one row each. We set `baseline_policy: null` on the other rows. ## Not staged (rules 8 and 9) -The rework dropped 1,267 of the first staging's 1,699 rows, and review A's fixes dropped 7 more (4 November), so 1,274 rows are not staged. The first staging's proposed keys that named only dropped rows are gone; their number can't be checked, because the first staging is in neither branch's history, so it is not given. +In the rework we dropped 1,267 of the first staging's 1,699 rows, and with review A's fixes we dropped 7 more (4 November), so we don't stage 1,274 rows. We retired the first staging's proposed keys that we had used only on dropped rows. We can't check how many there were, because neither branch's history includes the first staging, so we give no number. | Category | Rows | Examples (locator) | |---|---:|---| | A. Measure parameters: rates, thresholds, payment amounts, dates, durations; worked-example inputs | 829 | 817 former `policy_parameter` rows, e.g. "£13.21" per hour (para 3.18), "£2,500" surplus earnings threshold (Impact on households para 1.4), "£13,319" withdrawal threshold (explainer FAQ). Also the £100 Household Energy Payment amount (4 rows, e.g. para 2.48); the £108 current gas standing charge (para 3.9); the £2,000 rise in the full new State Pension over the Parliament (para 5.23); six worked-case inputs, e.g. Tom's "£10,300" income. | -| B. Spending allocations: programme funding, capitalisations, investment totals, spending levels | 221 | 166 programme amounts that are not a Table 6.1 line's printed impact (table above), e.g. "£120 million" for the AI Security Institute (para 4.130), "£30 million" for Cardiff Parkway (para 4.157), "£1.5 billion" for the Social and Affordable Homes Programme (para 5.4); 46 investment totals, e.g. NWF "£15.4 billion" (para 4.44), Your First Home "£4.7 billion" (para 5.5); public sector net investment (Table 1.2, 6); departmental spending levels (2); £200 million of private AI model credits (para 4.132). | +| B. Spending allocations: programme funding, capitalisations, investment totals, spending levels | 221 | 166 programme amounts that match no Table 6.1 line's printed impact (table above), e.g. "£120 million" for the AI Security Institute (para 4.130), "£30 million" for Cardiff Parkway (para 4.157), "£1.5 billion" for the Social and Affordable Homes Programme (para 5.4); 46 investment totals, e.g. NWF "£15.4 billion" (para 4.44), Your First Home "£4.7 billion" (para 5.5); public sector net investment (Table 1.2, 6); departmental spending levels (2); £200 million of private AI model credits (para 4.132). | | C. Financing: DMO remit, gilts, Treasury bills, NS&I, cash requirements, financial transactions | 177 | DMO net financing requirement "£258.6 billion" (Annex A.8); Table A.1 gilt sales by maturity; CGNCR "£144.1 billion" (para 1.47); Table 1.4 financial transactions by programme, e.g. Student Loans "15.9" (55 rows); Sovereign AI "£60 million" (para 4.135). | | D. Contingent liabilities | 14 | Table 1.5, e.g. Department for Energy Security and Net Zero "1.2-1.7"; Total reasonable worst case "4-4.5". | | E. Programme inputs, scope counts and targets | 25 | "670" HMRC compliance officers to recruit (paras 2.42, 6.151); tariffs on "over 100" food products (para 3.17); "around 10,000" homes the Social and Affordable Homes Programme funds (para 6.114); the 25% administrative-burden target worth "£5.6 billion" (para 4.51); "over 90% of England's population will have a devolution agreement in place" (para 4.68), a coverage count that includes existing agreements (review A's RB-7, dropped 4 November). | -| F. Effects of policies outside this Budget | 3 | "more than 100" businesses using the Global Talent funder pathway since Budget 2025 (para 4.112); "over £5 billion" of services exports from the Switzerland and South Korea deals (para 4.124); "over £2 billion of net progress secured" against the administrative-burden target (para 4.51), progress before this Budget (footnote 137 cites a July 2026 annex; para 4.52 counts this Budget's £640 million separately; RB-7). | +| F. Effects of policies outside this Budget | 3 | "more than 100" businesses using the Global Talent funder pathway since Budget 2025 (para 4.112); "over £5 billion" of services exports from the Switzerland and South Korea deals (para 4.124); "over £2 billion of net progress secured" against the administrative-burden target (para 4.51), progress before this Budget (in footnote 137 HM Treasury cites a July 2026 annex, and in para 4.52 it counts this Budget's £640 million separately; RB-7). | | G. Repeats beyond two of one claim | 4 | The Household Energy Payment's "£2.6 billion" in para 3.5 and Box 3.A, and "up to 20 million households" in para 3.3 and Box 3.A. Kept: the £2.6 billion of the executive summary and para 2.48, and the 20 million of the executive summary and paras 2.48 and 3.5 (review A's RB-2; see Review A fixes). | -| H. The rework brief's drop list | 1 | Impact on households para 1.10's "over four times as much in public spending than they pay in tax" (`spending_to_tax_ratio`, which the brief's item 1 names for removal; RB-13). Row 365's 60% beside it stays as a printed value. | +| H. The rework brief's drop list | 1 | Impact on households para 1.10's "over four times as much in public spending than they pay in tax" (`spending_to_tax_ratio`, which the coordinator named for removal in the brief's item 1; RB-13). Row 359's 60% beside it stays as a printed value. | + +Under rule 9 we moved 74 kept rows from proposed keys to a Table 6.1 line's key: +- the 66 Table 6.3 item cells and the 160,000 employees (from proposed keys such as `ab2026__package_tax_gap_timely_payment`; now line 45 components); +- four food-tariff savings (now line 10 components); +- the £3 billion and £1.7 billion DIP parts (now line 41 components); +- the Your First Home count (now line 37). -Rule 9 moved 74 kept rows from proposed keys to a Table 6.1 line's key: the 66 Table 6.3 item cells and the 160,000 employees (proposed keys such as `ab2026__package_tax_gap_timely_payment`, now line 45 components); four food-tariff savings (now line 10 components); the £3 billion and £1.7 billion DIP parts (now line 41 components); and the Your First Home count (now line 37). One row stays off a line's key: the £1.4 billion of DWP fraud savings, which reconciles with no printed basis for line 47 (rule 3; see its proposed key). +We keep one row off a line's key: the £1.4 billion of DWP fraud savings, which reconciles with no printed basis for line 47 (rule 3; see its package key, `ab2026__package_dwp_fraud_and_error_measures`). ### Other material not staged -- **Tables 6.1, 6.2 and 6.4** belong to `uk_hmt_tables`. Table 6.3 is a separate printed breakdown and is staged here. -- **No Impact on households bar readings** (README rule 4). Figures 1.A–1.E print no data labels. Chart 1.6's six data labels are staged. -- **Words without numbers** ("hundreds of pounds a month", "thousands of jobs") are not turned into numbers. -- **Footnote 73's earlier-event yields** (£6.5 billion, over £1 billion, £2.4 billion) are other fiscal events' measures; only this Budget's £2.3 billion and the £12 billion total are staged. -- **Background macro forecasts** that score no measure (GDP growth, inflation, employment, Table 1.1), departmental efficiency histories, Annex B budgets and Annex C charts are not staged. -- **The abbreviated 29 October explainer** has no quantitative content. -- **Other figures not staged (review A's RB-20(c)):** paragraph 5.17's Motability saving "over £1 billion over the next five years" (no Table 6.1 line); paragraph 3.7's £150 from Budget 2025 (an earlier event's measure); paragraph 4.36's reprioritisation savings (spending allocations); paragraph 4.42's National Wealth Fund and British Business Bank outcomes to date (effects before this Budget); and paragraph 6.9's £150 million (an allocation). -- **Paragraph 1.38's** repeat of the OBR's 0.1 percentage point CPI effect (row 185 stages paragraph 3.2's). +- **Tables 6.1, 6.2 and 6.4:** we stage these in `uk_hmt_tables`. HM Treasury prints Table 6.3 as a separate breakdown, and we stage it here. +- **No Impact on households bar readings** (README rule 4). HM Treasury prints no data labels on Figures 1.A–1.E, and we stage Chart 1.6's six data labels. +- **Words without numbers:** we don't turn "hundreds of pounds a month" or "thousands of jobs" into numbers. +- **Footnote 73's earlier-event yields** (£6.5 billion, over £1 billion, £2.4 billion): HM Treasury gives them for other fiscal events' measures, and we stage only this Budget's £2.3 billion and the £12 billion total. +- **Background:** we don't stage macro forecasts that score no measure (GDP growth, inflation, employment, Table 1.1), departmental efficiency histories, Annex B budgets or Annex C charts. +- **The abbreviated 29 October explainer:** HM Treasury gives no quantitative content in it. +- **Other figures not staged (review A's RB-20(c)):** + - paragraph 5.17's Motability saving "over £1 billion over the next five years" (no Table 6.1 line); + - paragraph 3.7's £150 from Budget 2025 (an earlier event's measure); + - paragraph 4.36's reprioritisation savings (spending allocations); + - paragraph 4.42's National Wealth Fund and British Business Bank outcomes to date (effects before this Budget); + - paragraph 6.9's £150 million (an allocation). +- **Paragraph 1.38:** HM Treasury repeats the OBR's 0.1 percentage point CPI effect there; in row 185 we stage paragraph 3.2's. ## Judgement calls for the coordinator -- **`spending_to_tax_ratio` is dropped (review A's RB-13).** The 3 November finish kept paragraph 1.10's "over four times" as a distributional statement. Review A read the rework brief's item 1 as a named removal list whose only exception is a Table 6.1 cost or yield: neither primary prints another spending-to-tax ratio, so the list can only mean this row, and HM Treasury frames paragraph 1.10 as "context beyond the impact of recent government policy decisions". The row is dropped (category H). The 60% row beside it, Table 2.C and the other Impact on households context rows stay as printed values. -- **Mislabelled shares beyond the 84%.** Besides the 84% (`unaffected_share`, never 16%), BADR's "three-quarters … will still receive relief on their entire gain" is the unaffected share too; "half of all RHL properties … will not pay business rates at all" includes properties already paying none (`share_with_zero_liability`); and Impact on households' 20%, 60%, 59% and approximately 20% are not affected shares (proposals above). The registered `share_no_change` is a household-income measure, so the 84% of income tax payers follows the brief's `unaffected_share`. -- **Registered metrics replace AB2025-era proposals.** The four "around £50 off the annual Ofgem price cap" rows and the three energy-debt "£30" rows now use the registered `energy_bill_change` (unit `gbp_per_household`), which the enum mints for HM Treasury bill figures; the India agreement's GDP effects use `gdp_level_effect` in GBP (the AB2025 IPPR precedent). -- **Financial transactions and public sector net investment are dropped as financing and spending levels.** Table 1.4 prints forecast levels of lending by programme, not Budget effects. Public sector net investment is an OBR aggregate, but the brief lists `public_investment_share` for removal. -- **Jobs and induced homes stay; delivery targets go.** Estimated jobs (NWF 100,000; BBB 40,000; Rolls-Royce 3,300; DIP 60,000; proposed `jobs_created_or_supported` since review A) and Your First Home's "estimated 20,000 new homes" are estimated effects. The 10,000 homes the Social and Affordable Homes Programme funds (in line 35's title), robotic procedures rising to 500,000, 2,000 learners and the 650,000 workforce requirement of the infrastructure pipeline are delivery targets of allocations. -- **The BBB's 40,000 jobs is now `restated`.** Footnote 135 cites the British Business Bank's press release. The NWF's 100,000 is `restated` too (footnote 134 cites the National Wealth Fund's own statement). +- **We dropped `spending_to_tax_ratio` (review A's RB-13).** In the 3 November finish we kept paragraph 1.10's "over four times" as a distributional statement. Review A read the rework brief's item 1 as a named removal list whose only exception is a Table 6.1 cost or yield. HM Treasury prints no other spending-to-tax ratio in either primary, so we read the removal list as applying only to this row. In paragraph 1.10 HM Treasury writes "context beyond the impact of recent government policy decisions". We dropped the row (category H). We keep the 60% row beside it, Table 2.C and the other Impact on households context rows as printed values. +- **Mislabelled shares beyond the 84%.** We stage the 84% as `unaffected_share`, never 16%. In BADR's "three-quarters … will still receive relief on their entire gain" HM Treasury gives the unaffected share too. In "half of all RHL properties … will not pay business rates at all" it counts properties already paying none (`share_with_zero_liability`). We don't read Impact on households' 20%, 60%, 59% and approximately 20% as affected shares (proposals above). We define `share_no_change` in the registry as a household-income measure, so for the 84% of income tax payers we follow the brief's `unaffected_share`. +- **Registered metrics replace AB2025-era proposals.** For the four "around £50 off the annual Ofgem price cap" rows and the three energy-debt "£30" rows we now use the registered `energy_bill_change` (unit `gbp_per_household`), which the enum mints for HM Treasury bill figures. For the India agreement's GDP effects we use `gdp_level_effect` in GBP (the AB2025 IPPR precedent). +- **We dropped financial transactions and public sector net investment as financing and spending levels.** In Table 1.4 HM Treasury prints forecast levels of lending by programme, not Budget effects. The OBR produces public sector net investment as an aggregate, but the coordinator listed `public_investment_share` for removal in the brief. +- **We keep jobs and induced homes, and drop delivery targets.** We stage the estimated jobs (NWF 100,000; BBB 40,000; Rolls-Royce 3,300; DIP 60,000; proposed `jobs_created_or_supported` since review A) and Your First Home's "estimated 20,000 new homes" as estimated effects. We treat the 10,000 homes the Social and Affordable Homes Programme funds (in line 35's title), robotic procedures rising to 500,000, 2,000 learners and the infrastructure pipeline's 650,000 workforce requirement as delivery targets of allocations. +- **We now mark the BBB's 40,000 jobs `restated`.** In footnote 135 HM Treasury cites the British Business Bank's press release. We mark the NWF's 100,000 `restated` too (in footnote 134 HM Treasury cites the National Wealth Fund's own statement). ## Printed problems and differences - **Spending figures that differ from Table 6.1:** - Social and Affordable Homes Programme: £1.5 billion (paras 5.4, 6.114) against line 35's -1,750 over 2027-28 to 2030-31. - - Early years books: "£5 million a year for three years from 2027-28" (paras 3.20, 6.122) against line 9's -5 in 2026-27, 2027-28 and 2028-29. The amount agrees; the years do not. - - Local apprenticeship services £100 million (line 22: -120); Good Growth Funds £285 million in 2027-28 (line 20: -340); Strategic Authorities £150 million (line 25: -180); Union Learning Fund £15 million a year (line 73: -20 a year); Sovereign AI £60 million (Table 6.2 line 4: -45 a year). + - Early years books: "£5 million a year for three years from 2027-28" (paras 3.20, 6.122) against line 9's -5 in 2026-27, 2027-28 and 2028-29. The amounts agree; the years don't. + - Local apprenticeship services £100 million (line 22: -120); + - Good Growth Funds £285 million in 2027-28 (line 20: -340); + - Strategic Authorities £150 million (line 25: -180); + - Union Learning Fund £15 million a year (line 73: -20 a year); + - Sovereign AI £60 million (Table 6.2 line 4: -45 a year). - DWP fraud and error: "an additional £1.4 billion over the forecast" (para 5.17) against line 47's +600m in total (+940m with Table 6.2 line 9; the Policy Costings' +570m). -- **Regulatory savings target:** paragraph 6.36 prints "£5.6 million"; paragraph 4.51 prints "£5.6 billion". The executive summary's "almost £800 million per year" and paragraph 4.52's "up to … £640 million per annum" have different printed scopes; both stay. -- **Repeated amounts that disagree (not staged after rule 8):** local AI partnerships £30 million (para 4.135) and £15 million (para 6.101); Leeds College of Building £28 million (para 4.71) and £15 million (para 6.53); business tariff suspensions on 269 goods (para 3.28), 350 goods (para 4.147) and 361 extensions plus 72 new suspensions (line 16's title). -- **Pubs relief:** the text prints an extra "20%"; Table 6.1 prints "20ppts". -- **Table 1.3** labels the new forecast "November 2026 forecast" in an October Budget. Chart 1.6 prints 14.5 where the text says £15 billion. "Debt is lower at the end of the forecast" fits public sector net financial liabilities (83.2% to 81.7%), while public sector net debt goes from 95.5% to 95.6%. -- **Paragraphs 6.225 and 6.226** repeat the Machine Games Duty measure. Annex A footnote 181 points to A.17 for green gilts, which A.19 covers. -- **Pension Tax Allowance explainer:** the FAQ prints "£13.319" for £13,319; the explainer prints both £13,036.40 and the rounded £13,037, and "up to £94" beside Steph's £93.28. The 28 October version's FAQ repeats a question and prints "part-time wok"; the version HM Treasury republished on 29 October (`83f8c894…`) fixes both and still prints "£13.319". +- **Regulatory savings target:** HM Treasury prints "£5.6 million" in paragraph 6.36 and "£5.6 billion" in paragraph 4.51. It gives different printed scopes to the executive summary's "almost £800 million per year" and paragraph 4.52's "up to … £640 million per annum", and we keep both. +- **Repeated amounts that disagree (not staged after rule 8):** + - local AI partnerships £30 million (para 4.135) and £15 million (para 6.101); + - Leeds College of Building £28 million (para 4.71) and £15 million (para 6.53); + - business tariff suspensions on 269 goods (para 3.28), 350 goods (para 4.147) and 361 extensions plus 72 new suspensions (line 16's title). +- **Pubs relief:** HM Treasury writes the extra relief as "20%" in the text and as "20ppts" in Table 6.1. +- **Table 1.3:** HM Treasury labels the new forecast "November 2026 forecast" in an October Budget. In Chart 1.6 it prints 14.5 where it writes £15 billion in the text. Its "Debt is lower at the end of the forecast" fits public sector net financial liabilities (83.2% to 81.7%), while public sector net debt goes from 95.5% to 95.6%. +- **Paragraphs 6.225 and 6.226:** HM Treasury repeats the Machine Games Duty measure. In Annex A footnote 181 it points to A.17 for green gilts, which A.19 covers. +- **Pension Tax Allowance explainer:** in the FAQ HM Treasury prints "£13.319" for £13,319. It prints both £13,036.40 and the rounded £13,037, and "up to £94" beside Steph's £93.28. In the 28 October version's FAQ it repeats a question and prints "part-time wok". In the version it republished on 29 October (`83f8c894…`) it fixes both and still prints "£13.319". ## Verification -- **Checker:** `PYTHONPATH=. uv run -q python pipeline/ab2026_harvest_contract.py sources/harvest-uk-ab2026-2026-10-29/uk_hmt_redbook` reports 425 rows, 0 problems (4 November, after review A's fixes, with the coordinator's tightened checks). -- **Determinism:** two runs of `stage.py` write the same `claims_staged.jsonl.gz` and an unchanged `manifest.jsonl`. COUNTS.json pins the gzip's sha256. -- **Spot check (the rework's, 29 October; row numbers are the rework's, which held at 4f424b9):** 30 rows drawn with `random.Random(20261029).sample(range(432), 30)` (0-based row numbers), each re-opened at its locator in the hashed primaries: red-book and Impact on households pages extracted with `pdftotext -layout -f N -l N`; table cells read against the printed row and column. All 30 match on value, sign, unit and period. +- **Checker.** `PYTHONPATH=. uv run -q python pipeline/ab2026_harvest_contract.py sources/harvest-uk-ab2026-2026-10-29/uk_hmt_redbook` reports 425 rows, 0 problems (4 November, after review A's fixes, with the coordinator's tightened checks). +- **Determinism.** Two runs of `stage.py` write the same `claims_staged.jsonl.gz` and an unchanged `manifest.jsonl`. COUNTS.json pins the gzip's sha256. +- **Spot check (the rework's, 29 October; row numbers are the rework's, which held at 4f424b9).** The rework drew 30 rows with `random.Random(20261029).sample(range(432), 30)` (0-based row numbers) and re-opened each at its locator in the hashed primaries. It extracted red-book and Impact on households pages with `pdftotext -layout -f N -l N`, and read table cells against the printed row and column. All 30 match on value, sign, unit and period. | Row | Locator | Printed | Staged value and unit | Period | |---:|---|---|---|---| @@ -304,54 +355,59 @@ Rule 9 moved 74 kept rows from proposed keys to a Table 6.1 line's key: the 66 T ### Tranche 1 finish (3 November 2026) -The rework stopped before its report, so on 3 November each instruction of `briefs/rework-hmt-redbook.md` (on the staging branch) was checked against the rows: +The rework stopped before its report, so on 3 November we checked each instruction of `briefs/rework-hmt-redbook.md` (on the staging branch) against the rows: -- **Removed metrics.** No `policy_parameter` row remains, and of the nine allocation and financing metrics only `spending_to_tax_ratio` remains, on one row (paragraph 1.10's "over four times"). It stays: it is a distributional statement printed in Impact on households, which item 3 of the brief keeps, not the fiscal aggregate the drop list targets. See the judgement calls above. -- **Rule 9 and the "Also fix" section.** The 84% row is `unaffected_share` as printed. No row carries `ab2026__package_tax_gap_timely_payment`; line 45's items carry line 45's key as components. No measure without a Table 6.1 line is named as a package. No kept row has the scope of the coordinator's shared spellings (see "Proposed measure keys"). -- **`component`.** Six rows still used `component` for a chart bar; they now use `chart_bar`. -- **Stale files.** The family had no REPORT.md or spot-check files left. -- **Keys** were harmonised across families on 3 November; `stage.py`'s `KEY_MAP` lists each rename, and the proposed keys above give the final spellings. -- **Checker:** 0 problems. **Determinism:** a re-run of `stage.py` reproduces the gzip and manifest byte for byte. -- **Spot check: not repeated.** The brief asked for 15 random rows re-opened at their locators. The 3 November session's network policy refused gov.uk and assets.publishing.service.gov.uk (HTTP 403 at the proxy), so no primary could be fetched, and no row was re-opened. The 30-row check above is the rework's own; nobody independent has checked this rework yet. +- **Removed metrics.** No `policy_parameter` row remained. Of the nine allocation and financing metrics, only `spending_to_tax_ratio` remained, on one row (paragraph 1.10's "over four times"). We kept it then as a distributional statement that HM Treasury prints in Impact on households, which we keep under item 3 of the brief, not as the fiscal aggregate we drop under the list. See the judgement calls above. +- **Rule 9 and the "Also fix" section.** We stage the 84% row as `unaffected_share`, as printed. We key no row to `ab2026__package_tax_gap_timely_payment`, and key line 45's items to line 45 as components. We name no measure without a Table 6.1 line as a package. No kept row has the scope of the coordinator's shared spellings (see "Keys we proposed"). +- **`component`.** Six rows still used `component` for a chart bar, and we moved them to `chart_bar`. +- **Stale files.** We found no REPORT.md or spot-check files left in the family. +- **Keys.** The coordinator harmonised keys across families on 3 November. We record each rename in `stage.py`'s `KEY_MAP` and give the final spellings under "Keys we proposed" above. +- **Checker:** 0 problems. **Determinism:** when we re-ran `stage.py` it reproduced the gzip and manifest byte for byte. +- **Spot check: not repeated.** The brief asked for 15 random rows re-opened at their locators. The 3 November session's network policy refused gov.uk and assets.publishing.service.gov.uk (HTTP 403 at the proxy), so we fetched no primary and re-opened no row. The 30-row check above is the rework's own; nobody independent has checked this rework yet. ## Key harmonisation, 3 November 2026 -The coordinator gave each scope one key across families (tranche 1, task 3.4). This family's renames: +The coordinator gave each scope one key across families (tranche 1, task 3.4), and renamed these keys in this family: - `ab2026__package_obr_forecast_policy_effects` → `ab2026__package_government_decisions_since_march_2026` - `ab2026__package_budget_barnett_consequentials` → `ab2026__package_barnett_consequentials` - `ab2026__defence_investment_plan` → `ab2026__package_defence_investment_plan` -The OBR's Table 3.1 "Total effect of Government decisions" prints the same figures as Table 1.3's "Total effect of policy decisions" (2.1 in 2028-29 in both), so the red book's restatement shares the OBR family's key. The Barnett key names the scope the Welsh Government's figures share (£450 million here, £448m there). The Defence Investment Plan crosses Table 6.1 lines 40–42, so it is a package, shared with the OBR's DIP funding row. Table 1.3's Tax and Spending parts stay components of the total, as staged. +The OBR prints the same figures in its Table 3.1 "Total effect of Government decisions" as HM Treasury does in Table 1.3's "Total effect of policy decisions" (2.1 in 2028-29 in both). So the red book's restatement shares the OBR family's key. We name the Barnett key for the scope the Welsh Government's figures share (£450 million here, £448m there). The Defence Investment Plan crosses Table 6.1 lines 40–42, so we key it as a package, shared with the OBR's DIP funding row. We keep Table 1.3's Tax and Spending parts as components of the total, as staged. ## README rule 10, 3 November 2026 -Table 1.3's total effect of policy decisions and its tax and spending parts (18 rows) already took `decisions_effect_on_borrowing`, and its two memo lines on the current budget deficit (11 rows) already proposed `fiscal_balance_change`. The memo rows' sign convention named PSNB, so `stage.py` now states which way the current budget deficit moves on rows 138–142 (underlying forecast differences) and 161–166 (policy decisions). No value or sign changed. The contract checker reports 432 rows, 0 problems, and a re-run of `stage.py` reproduces the gzip byte for byte. +We had already put Table 1.3's total effect of policy decisions and its tax and spending parts (18 rows) on `decisions_effect_on_borrowing`, and proposed `fiscal_balance_change` for its two memo lines on the current budget deficit (11 rows). In the memo rows' sign convention we had named PSNB, so in `stage.py` we now state which way the current budget deficit moves on rows 138–142 (underlying forecast differences) and 161–166 (policy decisions). We changed no value or sign. The contract checker reports 432 rows, 0 problems, and when we re-run `stage.py` it reproduces the gzip byte for byte. ## Review A fixes, 4 November 2026 -Review A (issue comment 6635870934 on #231, at 4f424b9) approved this family with fixes. `stage.py` applies each as an explicit, commented step after the rows, indexed by 4f424b9's row numbers, and drops rows only at the end. Row numbers in this section are 4f424b9's too, as in the review; the seven dropped rows move every later row up to seven places earlier in the staged file (row 365 here is row 359 there). No primary was re-opened for this round; every new quote and value comes from the review or the row. The coordinator then finished RB-18 and RB-19 from two pinned primaries (below). 432 rows → 425; `own` 263 → 257, `restated` 169 → 168; 6 hashed primaries. - -- **RB-1:** paragraph 5.17's two totals "over the forecast" (rows 265 and 266) sit at 2031-32 with the forecast's horizon; row 265 is `cumulative`, row 266 keeps `range_low`; row 267 is `point_in_time`. -- **RB-2:** rows 191 and 200 dropped (the £2.6 billion's third and fourth printings); row 179 dated 2026-27. The same rule drops rows 188 and 201, which repeat "up to 20 million" in paragraph 3.3 and Box 3.A: once RB-11 put rows 180 and 201 in England, four rows shared one claim identity (the coordinator's new duplicate-claim check, H-3, rejects more than two). -- **RB-3:** row 226 moved to `ab2026__package_pubs_relief_35pc_total`. -- **RB-4:** Chart 1.6's bridge steps (rows 21, 22, 24) propose `fiscal_balance_change` with the stability rule's sign; row 24 carries the government-decisions key; all six rows name the stability rule, and rows 20, 23 and 25 their forecast vintage and policy scenario. -- **RB-5:** row 246 is a line 29 component (its proposed key is retired); rows 232–237 are line 14 components; row 254 is a line 77 component. -- **RB-6:** rows 219–220's component says it contains rows 217–218. -- **RB-7:** rows 247 and 252 dropped (the coordinator's call), with their one-row keys `ab2026__package_parliament_regulatory_burden_reduction` and `ab2026__english_devolution_agreements`; categories F and E list them. -- **RB-8:** the DWP fraud key's scope and row 265's note rewritten; rows 265–266 `hmt_published_estimate_method_unspecified`. -- **RB-9:** rows 245, 246, 253 and 268 propose `jobs_created_or_supported`; row 245 is `restated`; rows 253 and 268 `hmt_published_estimate_method_unspecified`. -- **RB-10:** rows 11 and 240 dated 2027-28 ("per year from 2027-28"); rows 6 and 11 `hmt_costing_obr_certified`. -- **RB-11:** geography corrected on rows 5, 8, 9, 14, 180, 190, 196–199, 201–205, 258, 269 and 275, and "Great Britain" written `GB` throughout (the harvest's closed list). -- **RB-12:** rows 30–39's sign conventions; rows 30, 31, 34, 35, 36 and 38 are ONS outturns (`administrative_data`, `administrative_fact`). -- **RB-13:** row 366 dropped (category H). -- **RB-14:** the republished explainer is a 0-row primary (manifest line 6). -- **RB-15:** repeats share a dating (rows 9, 10, 14, 196, 205, 248–251, 271, 273), and rows 0, 19, 215 and 244 take the multi-year convention. -- **RB-16:** rows 143–166's `reform_hint` is "Total effect of policy decisions". -- **RB-17:** rows 185, 216, 226 and 237 have conventions `signs.py` reads; row 185 keeps its key, with the three scopes in its bullet; row 182 is 0.84 `share`. -- **RB-18:** rows 33 and 37 are HM Treasury's own arithmetic; rows 6, 7, 8, 10, 11, 12 and 14 take their repeats' source models (row 5's and row 9's repeats are already `arithmetic`; row 13 has no repeat and keeps its label); row 222 is a worked example (`arithmetic`, with its subgroup); row 183 `decile_10`; row 184 dated 2027-28, medium confidence, `hmt_published_estimate_method_unspecified`; row 168 loses `presentation`; row 267 `hmt_published_estimate_method_unspecified`. Row 194's note gives footnote 84's "fuel stressed" definition, which the coordinator read from the red book (19b337ac…). -- **RB-19:** rows 15, 29, 41 and 42 quote only their sentence; rows 103, 104 and 167 lose their leading paragraph numbers; locators of rows 172, 190, 227, 248 and 249 corrected. Rows 365 and 368 quote their whole sentences, which the coordinator read from Impact on households (36cd5afb…); row 365's sentence includes the lead-in that dates para 1.10's bullets to 2028-29 (row 366 is dropped). -- **RB-20:** (a) the unverifiable "234 proposed keys" figure is removed; (b) the spending test reads "the per-year amount equals the line's", which rows 224 and 276 pass, so they stay; (c) five more figures listed under "Other material not staged". -- **RB-21:** this lane folder's gzip and NOTES now match #231's; the 40 proposed-metric rows set `metric: null` explicitly. Table 6.3's duplication of Table 6.1 lines 45 and 46 (`uk_hmt_tables`) is left to tranche 2's cross-family dedupe. -- **K-7(b):** row 258 is the Barnett `total`, and rows 258–261 carry the sign convention. -- **K-7(c):** row 222 already proposes `gross_earnings_change`, the AB2025 spelling; no change. +Review A (issue comment 6635870934 on #231, at 4f424b9) approved this family with fixes. `stage.py` applies each as an explicit, commented step after the rows, indexed by 4f424b9's row numbers, and drops rows only at the end. We give 4f424b9's row numbers in this section too, as in the review. By dropping seven rows we move every later row up to seven places earlier in the staged file (row 365 here becomes row 359 there). We re-opened no primary for this round; every new quote and value comes from the review or the row. The coordinator then finished RB-18 and RB-19 from two pinned primaries (below). 432 rows → 425; `own` 263 → 257, `restated` 169 → 168; 6 hashed primaries. + +- **RB-1:** we put paragraph 5.17's two totals "over the forecast" (rows 265 and 266) at 2031-32 with the forecast's horizon. We marked row 265 `cumulative`, kept `range_low` on row 266 and marked row 267 `point_in_time`. +- **RB-2:** we dropped rows 191 and 200 (the £2.6 billion's third and fourth printings) and dated row 179 2026-27. Under the same rule we dropped rows 188 and 201, which repeat "up to 20 million" in paragraph 3.3 and Box 3.A. Once RB-11 put rows 180 and 201 in England, four rows shared one claim identity, and the coordinator's new duplicate-claim check, H-3, rejects more than two. +- **RB-3:** we moved row 226 to `ab2026__package_pubs_relief_35pc_total`. +- **RB-4:** for Chart 1.6's bridge steps (rows 21, 22, 24) we propose `fiscal_balance_change` with the stability rule's sign, and key row 24 to the government-decisions key. On all six rows we name the stability rule, and on rows 20, 23 and 25 their forecast vintage and policy scenario. +- **RB-5:** we made row 246 a line 29 component (retiring its proposed key), rows 232–237 line 14 components and row 254 a line 77 component. +- **RB-6:** in rows 219–220's component we say it contains rows 217–218. +- **RB-7:** we dropped rows 247 and 252 (the coordinator's call), with their one-row keys `ab2026__package_parliament_regulatory_burden_reduction` and `ab2026__english_devolution_agreements`, and list them under categories F and E. +- **RB-8:** we rewrote the DWP fraud key's scope and row 265's note, and marked rows 265–266 `hmt_published_estimate_method_unspecified`. +- **RB-9:** we propose `jobs_created_or_supported` for rows 245, 246, 253 and 268, mark row 245 `restated`, and mark rows 253 and 268 `hmt_published_estimate_method_unspecified`. +- **RB-10:** we dated rows 11 and 240 2027-28 ("per year from 2027-28"), and marked rows 6 and 11 `hmt_costing_obr_certified`. +- **RB-11:** we corrected the geography on rows 5, 8, 9, 14, 180, 190, 196–199, 201–205, 258, 269 and 275, and write "Great Britain" as `GB` throughout (the harvest's closed list). +- **RB-12:** we fixed rows 30–39's sign conventions, and marked rows 30, 31, 34, 35, 36 and 38 as ONS outturns (`administrative_data`, `administrative_fact`). +- **RB-13:** we dropped row 366 (category H). +- **RB-14:** we record the republished explainer as a 0-row primary (manifest line 6). +- **RB-15:** we gave repeats one dating (rows 9, 10, 14, 196, 205, 248–251, 271, 273), and moved rows 0, 19, 215 and 244 to the multi-year convention. +- **RB-16:** we set rows 143–166's `reform_hint` to "Total effect of policy decisions". +- **RB-17:** we gave rows 185, 216, 226 and 237 conventions `signs.py` reads. We keep row 185's key, with the three scopes in its bullet, and stage row 182 as 0.84 `share`. +- **RB-18:** + - We mark rows 33 and 37 as HM Treasury's own arithmetic. + - We give rows 6, 7, 8, 10, 11, 12 and 14 their repeats' source models. We already used `arithmetic` for row 5's and row 9's repeats, and keep row 13's label without a repeat to follow. + - We stage row 222 as a worked example (`arithmetic`, with its subgroup), and row 183 as `decile_10`. + - We dated row 184 2027-28, at medium confidence, `hmt_published_estimate_method_unspecified`. We removed `presentation` from row 168, and marked row 267 `hmt_published_estimate_method_unspecified`. + - In row 194's note we give footnote 84's "fuel stressed" definition, which the coordinator read from the red book (19b337ac…). +- **RB-19:** in rows 15, 29, 41 and 42 we now quote only the sentence, and from rows 103, 104 and 167 we drop the leading paragraph numbers. We corrected the locators of rows 172, 190, 227, 248 and 249. In rows 365 and 368 we quote whole sentences, which the coordinator read from Impact on households (36cd5afb…). In row 365's sentence HM Treasury dates para 1.10's bullets to 2028-29 in its lead-in (we dropped row 366). +- **RB-20:** (a) We removed the unverifiable "234 proposed keys" figure. (b) Our spending test now reads "the per-year amount equals the line's", which rows 224 and 276 pass, so we keep them. (c) We list five more figures under "Other material not staged". +- **RB-21:** we matched this lane folder's gzip and NOTES to #231's, and set `metric: null` explicitly on the 40 proposed-metric rows. We left Table 6.3's duplication of Table 6.1 lines 45 and 46 (`uk_hmt_tables`) to tranche 2. At ingest we drop rows 339–350, which repeat those lines, as twins of `uk_hmt_tables` rows 264–275, and record each under the kept claim's `publication.also_printed_in`. +- **K-7(b):** we made row 258 the Barnett `total`, and gave rows 258–261 the sign convention. +- **K-7(c):** row 222 already proposes `gross_earnings_change`, the AB2025 spelling, so we made no change. diff --git a/sources/harvest-uk-ab2026-2026-10-29/uk_hmt_tables/NOTES.md b/sources/harvest-uk-ab2026-2026-10-29/uk_hmt_tables/NOTES.md index 2173ce099..222eb7a5f 100644 --- a/sources/harvest-uk-ab2026-2026-10-29/uk_hmt_tables/NOTES.md +++ b/sources/harvest-uk-ab2026-2026-10-29/uk_hmt_tables/NOTES.md @@ -1,61 +1,61 @@ # uk_hmt_tables: HM Treasury Budget 2026 Tables 6.1, 6.2 and 6.4 -Staged 2026-10-29 by `pipeline/stage_ab2026_hmt_tables.py`, which is vendored, so the staging re-runs byte for byte. Source slug `hm_treasury`, source_model `hmt_costing_obr_certified`. +We staged this family on 2026-10-29 with the vendored `pipeline/stage_ab2026_hmt_tables.py`, so the staging re-runs byte for byte. We use source slug `hm_treasury` and source_model `hmt_costing_obr_certified`. -On 4 November 2026 the stager took review A's fixes T-1 to T-9. The workbooks weren't re-fetched in that session, so its output was produced by running the stager on the two workbooks' cells rebuilt in memory from the 29 October rows (every row's locator names its cell), the five section fills and Table 6.2's total row (cells E14:J14, as review A read them). Under the 29 October stager, that rebuild reproduces the 29 October claims file byte for byte. The rebuild script is on the staging branch (`harvest-staging/ab2026/tables-reconstruction/`); a run of the stager on the pinned workbooks should give this file's sha256. +On 4 November 2026 we applied review A's fixes T-1 to T-9 to the stager. We didn't re-fetch the workbooks in that session, so we rebuilt the two workbooks' cells in memory and ran the stager on them. We rebuilt them from the 29 October rows (we name each row's cell in its locator), the five section fills and Table 6.2's total row (cells E14:J14, as review A read them). Under the 29 October stager, that rebuild reproduces the 29 October claims file byte for byte. We keep the rebuild script on the staging branch (`harvest-staging/ab2026/tables-reconstruction/`); a run of the stager on the pinned workbooks should give this file's sha256. -- **Rows:** 768. +- Rows: 768. - Table 6.1: 98 measure lines and 3 total rows, each across 6 fiscal years (606 rows). - - Table 6.2: 9 lines and its total row, each across 6 fiscal years (60 rows). The total's 6 rows come last in the file, after Table 6.4's, so the rows staged before it keep their positions. + - Table 6.2: 9 lines and its total row, each across 6 fiscal years (60 rows). We put the total's 6 rows last in the file, after Table 6.4's, so the rows we staged before it keep their positions. - Table 6.4: 17 lines across 6 fiscal years (102 rows). -- **Primaries:** 2 workbooks. +- Primaries: 2 workbooks. ## Access recipe -The two XLSX attachments of [Supporting documents for Budget 2026](https://www.gov.uk/government/publications/supporting-documents-for-budget-2026), listed through the gov.uk content API (`/api/content/government/publications/supporting-documents-for-budget-2026`). - -- **First download:** 2026-10-28, `curl -sSL` with a browser User-Agent. -- **Second download:** 2026-10-29. Both hash to the pinned sha256, which the stager checks before writing anything. -- **Reading:** openpyxl with cached values (`data_only=True`), so every value is the number HM Treasury stored, not a recomputed formula. - -## What is staged - -- **Table 6.1:** every cell of every measure line, 2026-27 to 2031-32. - - Fields: `metric: revenue_change`, `unit_concept: gbp`, value = cell × 1e6. Since 3 November, the three total rows take `decisions_effect_on_borrowing` (README rule 10), with no value or sign changed. - - Conditions carry the line (`table_line`), the Tax/Spend head (`head`) and the line's headings (`heading`). - - `heading` is "section > sub-heading". The workbook shades its five top-level section headings (cells B5, B27, B43, B58 and B82, fill FFFFCCCC), as the red book shades them, and each sub-heading belongs to the last section above it. So lines 12–19 read "Breathing space for families and businesses > Breathing space for businesses". - - Column D of both sheets, headed "Fiscal Head check (hide)", is a hidden check column the red book doesn't print (15 lines read "SUBTOTAL", line 30 has "Tax" on a Spend line and line 44 "RAME" on a Tax line). It isn't staged (review A, T-3); `head` is the printed Tax/Spend head. - - Titles are staged with whitespace normalised: sixteen titles carry tabs, trailing spaces or no-break spaces in the workbook, collapsed to single spaces. Lines 45 and 47 print footnote marker 4 as a superscript run glued to the title; the stager strips it (review A, T-5). - - The three total rows carry the package keys `ab2026__package_total_policy_decisions`, `…_total_tax_policy_decisions` and `…_total_spending_policy_decisions`. - - The lines don't sum exactly to the totals: by up to £10m a year for the grand total and up to £15m for the Tax and Spend totals (the Spend total in 2029-30). Footnote 1 says totals may not sum due to rounding, and the rows are staged as printed. - - The totals' note quotes footnotes 1 and 2. Footnote 2 sets their scope: it excludes the PWLB debt relief and the landlord registration fee ("no net increase to borrowing"), and "AME and tax costings with an impact of less than +-£10 million in every year". -- **Table 6.2:** financial transactions, `proposed_metric: financial_transaction_cash_impact` (cash impact on the public sector, not borrowing). - - Each line carries the measure key of the Table 6.1 measure it belongs to. - - Line 4 (Sovereign AI) has no Table 6.1 line and so has its own registered key. - - Line 6, "Reprioritise financial transactions budgets to fund the new equity loan scheme", has `measure_key: null` and the proposed key `ab2026__your_first_home_ft_reprioritisation` (scope: Table 6.2 line 6; no Table 6.1 line), as Table 6.1 line 34's reprioritisation has its own line (review A, T-1). Line 5 is the scheme itself (its cells sum to −4,700, the red book's "£4.7 billion", para 5.5) and keeps `ab2026__your_first_home_equity_loan`. Both lines are whole rows, so they mustn't share a key: marking both as parts of the scheme would net it to an inflow. - - The total row, "Total cash impacts of financial transactions" (cells E14:J14: −230, −1,160, −2,290, −2,795, −3,675, −3,850), is staged as 6 `row_type: total` rows with `measure_key: null` and the proposed key `ab2026__package_total_financial_transactions` (scope: exactly Table 6.2 lines 1–9), with footnote 1's rounding note (review A, T-2). -- **Table 6.4:** recostings at Budget 2026 of measures announced earlier. - - Each row carries **that event's** registry key, with `conditions.recosting_of`: +HM Treasury attaches the two XLSX files to [Supporting documents for Budget 2026](https://www.gov.uk/government/publications/supporting-documents-for-budget-2026), and we listed them through the gov.uk content API (`/api/content/government/publications/supporting-documents-for-budget-2026`). + +- We first downloaded them on 2026-10-28, with `curl -sSL` and a browser User-Agent. +- We downloaded them again on 2026-10-29. Both downloads hash to the pinned sha256, which the stager checks before it writes anything. +- We read them with openpyxl's cached values (`data_only=True`), so we stage the number HM Treasury stored, not a recomputed formula. + +## What we staged + +- Table 6.1: every cell of every measure line, 2026-27 to 2031-32. + - We record `metric: revenue_change`, `unit_concept: gbp` and value = cell × 1e6. Since 3 November, under README rule 10, we give the three total rows `decisions_effect_on_borrowing` instead, with no value or sign changed. + - We record the line in `table_line`, the Tax/Spend head in `head` and the line's headings in `heading`. + - We write `heading` as "section > sub-heading". HM Treasury shades its five top-level section headings in the workbook (cells B5, B27, B43, B58 and B82, fill FFFFCCCC), as it does in the red book, and each sub-heading falls under the last section above it. So we write lines 12–19 as "Breathing space for families and businesses > Breathing space for businesses". + - HM Treasury hides a check column, headed "Fiscal Head check (hide)", in column D of both sheets, and doesn't print it in the red book. In that column HM Treasury enters "SUBTOTAL" on 15 lines, "Tax" on line 30, a Spend line, and "RAME" on line 44, a Tax line. We left the column out after review A's T-3, and take `head` from the printed Tax/Spend head. + - We normalise whitespace in titles: in sixteen titles HM Treasury typed tabs, trailing spaces or no-break spaces, and we collapse them to single spaces. On lines 45 and 47 HM Treasury prints footnote marker 4 as a superscript run glued to the title, and since review A's T-5 the stager strips it. + - We key the three total rows to the package keys `ab2026__package_total_policy_decisions`, `…_total_tax_policy_decisions` and `…_total_spending_policy_decisions`. + - The lines don't sum to the totals: they miss by up to £10m a year for the grand total and up to £15m for the Tax and Spend totals (the Spend total in 2029-30). In footnote 1 HM Treasury says totals may not sum due to rounding, and we staged the rows as printed. + - We quote footnotes 1 and 2 in the totals' note. In footnote 2, HM Treasury sets their scope: it excludes the PWLB debt relief and the landlord registration fee ("no net increase to borrowing"), and "AME and tax costings with an impact of less than +-£10 million in every year". +- Table 6.2: financial transactions, which we stage with `proposed_metric: financial_transaction_cash_impact` (cash impact on the public sector, not borrowing). + - We give each line the measure key of the Table 6.1 measure it belongs to. + - HM Treasury prints no Table 6.1 line for line 4 (Sovereign AI), so we gave it its own registered key. + - On line 6 HM Treasury prints "Reprioritise financial transactions budgets to fund the new equity loan scheme". We key it to `ab2026__your_first_home_ft_reprioritisation` (scope: Table 6.2 line 6; no Table 6.1 line), as Table 6.1 line 34's reprioritisation has its own line (review A, T-1). On 4 November we gave it `measure_key: null` and this key as a proposal, and in tranche 2 we registered it. HM Treasury puts the scheme itself on line 5 (its cells sum to −4,700, the red book's "£4.7 billion", para 5.5), and we keep `ab2026__your_first_home_equity_loan` on it. We stage both lines as whole rows, so they mustn't share a key: if we marked both as parts of the scheme, we would net it to an inflow. + - We stage the total row, "Total cash impacts of financial transactions" (cells E14:J14: −230, −1,160, −2,290, −2,795, −3,675, −3,850), as 6 `row_type: total` rows. We key them to `ab2026__package_total_financial_transactions` (scope: exactly Table 6.2 lines 1–9), which we proposed on 4 November and registered in tranche 2, and give them footnote 1's rounding note (review A, T-2). +- Table 6.4: HM Treasury's recostings at Budget 2026 of measures announced earlier. + - We key each row to **that event's** registry key, with `conditions.recosting_of`: - lines 1–14: `autumn_budget_2025` (`data/uk/ab2025_measures.json`); - lines 15–16: `autumn_budget_2024`; - line 17: `spring_budget_2024` (`data/uk/events/*_measures.json`). - - The cells are £ million. The table's units header prints "(£)", and its footnotes give no unit, but every value cell carries the number format `\+#,##0"m"` (I9 = 7130 displays as "+7,130m"), so the rows are `parse_confidence: high` (review A, T-6). The sheet is read by name ("Rounded Table 6.4"), not as the workbook's active sheet. - - Footnote 2 (cell F25) sets the table's scope: "Measures that interact with new measures announced at this Budget, for example on savings income tax changes, as well as measures that extend an existing policy such as personal tax threshold measures have not been included." (review A, T-6). + - HM Treasury reports the cells in £ million. It prints "(£)" in the units header and gives no unit in the footnotes, but formats every value cell as `\+#,##0"m"` (I9 = 7130 displays as "+7,130m"). We therefore give the rows `parse_confidence: high` (review A, T-6). We read the sheet by name ("Rounded Table 6.4"), not as the workbook's active sheet. + - In footnote 2 (cell F25), HM Treasury sets the table's scope: "Measures that interact with new measures announced at this Budget, for example on savings income tax changes, as well as measures that extend an existing policy such as personal tax threshold measures have not been included." (review A, T-6). -## Conventions (load-bearing) +## Conventions the rows depend on -- **Sign.** The tables print no legend. `conditions.sign_convention` on every Table 6.1 and 6.4 row reads "positive = yield to the Exchequer (reduces borrowing); negative = cost to the Exchequer". The convention is observed from the rows themselves: the bank surcharge rise is positive, the Household Energy Payment negative. The AB2025 Table 4.1 harvest uses the same convention. Values are NOT re-signed. -- **Table 6.2's sign.** Every Table 6.2 row reads "positive = improves the public sector's cash position (cash inflow or lower outflow); negative = worsens it". The National Wealth Fund's investment is negative; line 6's positives are a lower outflow (budgets reprioritised away from other financial transactions), not an inflow (review A, T-9). -- **scoring_method `final`.** Table 6.1 is the certified post-behavioural total. The static and behavioural legs are in the policy costings document (family `uk_hmt_costings`). -- **period** is the FY end year (2026-27 → 2027), with `conditions.fy` holding the label. -- **geography** is "UK" on every row: these are Exchequer impacts. A measure's own scope (for example England only for the Household Energy Payment) is in its title, verbatim in `reform_hint`, and in the registry. -- **quote** is the line's title and the cell, as printed (Open Government Licence). +- **Sign.** HM Treasury prints no legend for the tables. On every Table 6.1 and 6.4 row we set `conditions.sign_convention` to "positive = yield to the Exchequer (reduces borrowing); negative = cost to the Exchequer". We read the convention from the rows themselves: HM Treasury prints the bank surcharge rise as positive and the Household Energy Payment as negative. We use the same convention in the AB2025 Table 4.1 harvest. We didn't re-sign any value. +- **Table 6.2's sign.** On every Table 6.2 row we set "positive = improves the public sector's cash position (cash inflow or lower outflow); negative = worsens it". HM Treasury prints the National Wealth Fund's investment as negative, and on line 6 HM Treasury's positives record a lower outflow (budgets reprioritised away from other financial transactions), not an inflow (review A, T-9). +- **scoring_method `final`.** HM Treasury prints the certified post-behavioural total in Table 6.1, and the static and behavioural legs in the policy costings document (family `uk_hmt_costings`). +- **period.** We use the FY end year (2026-27 → 2027) and put the label in `conditions.fy`. +- **geography.** We set "UK" on every row, since HM Treasury reports Exchequer impacts. HM Treasury gives a measure's own scope (for example England only for the Household Energy Payment) in its title, which we copy verbatim into `reform_hint`, and we record it in the registry. +- **quote.** We quote the line's title and the cell, as HM Treasury prints them (Open Government Licence). -## For tranche 2 +## At ingest -- Table 6.2's proposed `financial_transaction_cash_impact` and the registered `cash_requirement_change`, which `uk_obr_tables` uses for paragraph 3.20's £1.9bn rise in the Government's financing requirement from the National Wealth Fund and British Business Bank transactions, both measure the cash effect of the same transactions: one as a cash impact, one as the financing requirement. The ingest settles the metric. -- Table 6.3 of the red book prints Table 6.1 lines 45 and 46 again (`uk_hmt_redbook` rows staged from it); tranche 2 dedupes identical HM Treasury figures across the two families. +- We stage Table 6.2 on the proposed `financial_transaction_cash_impact`, and at ingest we put its 60 rows on the registered `cash_requirement_change`, the PSNCR metric. In `uk_obr_tables` we stage paragraph 3.20's £1.9bn rise in the Government's financing requirement from the National Wealth Fund and British Business Bank transactions (row 2684) on the same metric. +- In Table 6.3 of the red book, HM Treasury prints Table 6.1 lines 45 and 46 again, and we staged `uk_hmt_redbook` rows 339–350 from it. At ingest we drop those 12 rows as twins of this family's rows 264–275 and record each under the kept claim's `publication.also_printed_in`. ## Not staged -- Table 6.1's footnote 2 names two measures with "no net increase to borrowing" that the table excludes: debt relief for councils in severe financial distress through the PWLB, and a landlord registration fee for the private rented sector. They have no figures, so there's nothing to stage. The registry doesn't list them. +- In Table 6.1's footnote 2, HM Treasury names two measures with "no net increase to borrowing" that it leaves out of the table. It names debt relief for councils in severe financial distress through the PWLB, and a landlord registration fee for the private rented sector. HM Treasury gives no figures for them, so we had nothing to stage, and our registry doesn't list them. diff --git a/sources/harvest-uk-ab2026-2026-10-29/uk_ifs_ab2026/NOTES.md b/sources/harvest-uk-ab2026-2026-10-29/uk_ifs_ab2026/NOTES.md index 82484ecce..495d32c92 100644 --- a/sources/harvest-uk-ab2026-2026-10-29/uk_ifs_ab2026/NOTES.md +++ b/sources/harvest-uk-ab2026-2026-10-29/uk_ifs_ab2026/NOTES.md @@ -1,16 +1,20 @@ # uk_ifs_ab2026 — Institute for Fiscal Studies, Autumn Budget 2026 -**Working files.** `stage.py`, which writes `claims_staged.jsonl.gz` and is the file meant wherever these notes say `stage.py`, lives on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/ifs/uk_ifs_ab2026/stage.py`; this branch carries only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz`. `stage.py` embeds every row, so it rebuilds the gzip with no network and no primary file. `.ab2026-primaries/uk_ifs_ab2026/` is the harvest's local primaries folder, which no commit carries: the repository holds no primary bytes. Fetch each deck from its `manifest.jsonl` URL, and each Flourish embed from the embed URL in its manifest note, and check its sha256. The two browser-text primaries are each page's `
` text as saved on 29 October; ifs.org.uk refuses curl and its HTML embeds per-request tokens, so only the saved texts match those two hashes. +We keep `stage.py`, which writes `claims_staged.jsonl.gz`, on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/ifs/uk_ifs_ab2026/stage.py`. We commit only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz` to this branch. We embedded every row in `stage.py`, so it rebuilds the gzip with no network and no primary file. We keep the local primaries in `.ab2026-primaries/uk_ifs_ab2026/` and never commit them. Fetch each deck from its `manifest.jsonl` URL, and each Flourish embed from the embed URL in its manifest note, and check its sha256. For the two browser-text primaries we took each page's `
` text as saved on 29 October. ifs.org.uk refuses curl and embeds per-request tokens in its HTML, so only the saved texts match those two hashes. -**144 rows from 13 primaries.** `source: ifs`. The manifest lists 14 primaries: four current briefing decks, the superseded TW_0 deck (no rows), two IFS pages read as browser text, and the public data of seven Flourish visualisations. The first lane staged 106 rows from the decks on 29 October. On 3 November the harvest applied the IFS addendum (`briefs/ifs-addendum.md` on the staging branch). Its sections 3 and 4 cut the deck rows to 87, through an explicit block that follows the deck rows in `stage.py`. Its sections 1 and 2 then appended 57 rows in a second JSON block: 11 from the two pages and 46 from the seven visualisations. Attribution: 48 `own`, 96 `restated`. No primary mentions PolicyEngine. From "Family conventions" on, the sections keep the first lane's record of 29 October; where they differ from the sections above them, the sections above win. +We stage **144 rows from 13 primaries**, with `source: ifs`. We list 14 primaries in the manifest: four current briefing decks, the superseded TW_0 deck (no rows), two IFS pages we read as browser text, and the public data of seven Flourish visualisations. +- The first lane staged 106 rows from the decks on 29 October. +- On 3 November we applied the IFS addendum (`briefs/ifs-addendum.md` on the staging branch). Under its sections 3 and 4 we cut the deck rows to 87, through an explicit block that follows the deck rows in `stage.py`. Under its sections 1 and 2 we then appended 57 rows in a second JSON block: 11 from the two pages and 46 from the seven visualisations. + +Attribution: 48 `own`, 96 `restated`. IFS mentions PolicyEngine in no primary. From "Family conventions" on, we keep the first lane's record of 29 October; where it differs from the sections above it, the sections above win. ## Review A fixes, 4 November 2026 -Review A (PR #231, issue comment 6635870934) asked for three changes to this family. They sit in a block before `main()` in `stage.py`; its indices are 0-based, as review A cites them. The coordinator read Adam's deck (slide 2) and Wernham's (slide 3) for K-7(a) and K-7(e); this session re-opened no primary. No printed value, sign or year changed. +Review A (PR #231, issue comment 6635870934) asked for three changes to this family. We put them in a block before `main()` in `stage.py`, with 0-based indices, as review A cites them. The coordinator read Adam's deck (slide 2) and Wernham's (slide 3) for K-7(a) and K-7(e); this session re-opened no primary. We changed no printed value, sign or year. -- **H-6(a):** rows 39 and 40 (`cpi_inflation_rate`) and 47 and 48 (`affected_share`) carry the registered metric in `metric`, with `proposed_metric` null. -- **K-7(a):** row 78 (£1.1bn, lines 61 and 62) keeps 2031-32 ("Biggest revenue-raisers in 2031–32") and adds `conditions.discrepancy`: "IFS prints £1.1bn for 2031-32, where Table 6.1 gives lines 61 and 62 a combined £1,045m." -- **K-7(e):** row 42 (£5bn): slide 3 gives it as the cost of the three measures "over next nine months", and Table 6.1 line 2 costs the whole VAT cut from 1 October 2026, so the re-key to lines 1–3 stands (£5,040m over 2026-27 and 2027-28). Its note says so, and it is `cumulative` with the time basis of RF's two-year totals: period 2028 (the final fiscal year), `fiscal_year`, no `fy`, and `conditions.horizon` naming both years. +- H-6(a): we now record the registered metric in `metric` on rows 39 and 40 (`cpi_inflation_rate`) and 47 and 48 (`affected_share`), with `proposed_metric` null. +- K-7(a): we keep row 78 (£1.1bn, lines 61 and 62) at 2031-32 ("Biggest revenue-raisers in 2031–32") and add `conditions.discrepancy`: "IFS prints £1.1bn for 2031-32, where Table 6.1 gives lines 61 and 62 a combined £1,045m." +- K-7(e): row 42 (£5bn). On slide 3 IFS gives it as the cost of the three measures "over next nine months", and HM Treasury's Table 6.1 line 2 costs the whole VAT cut from 1 October 2026. So the re-key to lines 1–3 stands (£5,040m over 2026-27 and 2027-28), and we say so in the row's note. We mark it `cumulative`, with the time basis of RF's two-year totals: period 2028 (the final fiscal year), `fiscal_year` and no `fy`, and we name both years in `conditions.horizon`. ## Coverage @@ -32,12 +36,12 @@ Review A (PR #231, issue comment 6635870934) asked for three changes to this fam | Flourish 30727239: higher and additional rate taxpayers | Flourish public data | 15 | 0 | 15 | | **Total** | | **144** | **48** | **96** | -Rows 0 to 86 come from the decks, by deck, slide and position. Rows 87 to 143 follow in read order: the initial response, the opening remarks, then the seven visualisations in the order above, each by data row and then series. +We take rows 0 to 86 from the decks, by deck, slide and position. Rows 87 to 143 follow in read order: the initial response, the opening remarks, then the seven visualisations in the order above, each by data row and then series. | measure_key | Rows | of which components (rule 9) | of which from sections 1 and 2 | |---|--:|--:|--:| -| `null`, context (no proposed key) | 116 | 0 | 52 | -| `null` with a proposed key | 7 | 0 | 3 | +| `null`, context (no key) | 116 | 0 | 52 | +| the four keys we proposed (below) | 7 | 0 | 3 | | `ab2026__household_energy_payment_100_bands_a_to_d` | 5 | 0 | 2 | | `ab2026__founders_relief_cgt` | 4 | 0 | 0 | | `ab2026__package_household_energy_vat_and_fuel_duty` | 3 | 0 | 0 | @@ -50,80 +54,95 @@ Rows 0 to 86 come from the decks, by deck, slide and position. Rows 87 to 143 fo | `ab2026__package_total_tax_policy_decisions` | 1 | 0 | 0 | | `ab2026__writing_down_allowance_12pc_and_fya_50pc` | 1 | 0 | 0 | -## Proposed measure keys +## Keys we proposed -Four proposed keys name seven rows. README rule 2 sets their spelling, and each takes the producer's printed scope (rule 3). +We proposed four keys for seven rows, and in tranche 2 we registered them and moved them into `measure_key`. We spell them under README rule 2 and give each the producer's printed scope (rule 3). -- `ab2026__package_property_and_savings_income_rates` (1 row): exactly Table 6.1 lines 61 and 62, Adam slide 2's "Increased income tax on savings & property income", £1.1bn in 2031–32 (lines 61 and 62 print 455 + 590 = 1,045). Slide 2's list opens "Biggest revenue-raisers in 2031–32", so the year stands, and the row's `conditions.discrepancy` records the £55m gap (review A, K-7(a)). -- `ab2026__gas_standing_charge_moved_to_unit_rates` (3 rows): no Table 6.1 line. Wernham slide 8's "Shifting gas standing charge onto unit price": "This charge will be scrapped, but per-unit price will be increased to absorb this". The spelling is the RF family's. -- `ab2026__package_ifs_new_giveaways` (2 rows, from the initial response): IFS's new giveaways, which push up borrowing by £8 billion this year and £4 billion next. The page names no OBR line and doesn't say the giveaways cover every Government decision since March, so the rows don't take the key the OBR, HM Treasury red book and RF families share for those decisions. To the nearest billion they match the OBR's direct effects of Government decisions in EFO Table 3.1, 7.6 in 2026-27 and 3.5 in 2027-28 (uk_obr_tables). We infer from that match that IFS restates the OBR, so both rows have medium confidence. Neither the total effect including indirect effects (6.3 and 7.0) nor Table 6.1's total of policy decisions (7,595 and 2,475 £m, uk_hmt_tables) matches both years. -- `ab2026__package_ifs_announced_tax_changes` (1 row, from Flourish 30726047): the tax change that Budget 2026 announced, at the forecast horizon, as a share of GDP as forecast at the event. The chart's footer credits IFS calculations from the OBR's October 2026 EFO, policy measures database and historical official forecasts database. The chart lists no Table 6.1 line and names no OBR grouping, and one rounded figure cannot tie it to the scope of `ab2026__package_total_tax_policy_decisions` or of the OBR's tax decisions since March, so the row proposes a key in IFS's own terms. +- `ab2026__package_property_and_savings_income_rates` (1 row): exactly Table 6.1 lines 61 and 62. On Adam slide 2 IFS gives "Increased income tax on savings & property income" as £1.1bn in 2031–32 (HM Treasury prints 455 + 590 = 1,045 on lines 61 and 62). IFS opens slide 2's list "Biggest revenue-raisers in 2031–32", so the year stands, and we record the £55m gap in the row's `conditions.discrepancy` (review A, K-7(a)). +- `ab2026__gas_standing_charge_moved_to_unit_rates` (3 rows): no Table 6.1 line. Wernham slide 8's "Shifting gas standing charge onto unit price": "This charge will be scrapped, but per-unit price will be increased to absorb this". We take the RF family's spelling. +- `ab2026__package_ifs_new_giveaways` (2 rows, from the initial response): IFS's new giveaways, which push up borrowing by £8 billion this year and £4 billion next. + - IFS names no OBR line and doesn't say the giveaways cover every Government decision since March. So we don't give the rows the key we share across the OBR, HM Treasury red book and RF families for those decisions. + - To the nearest billion they match the OBR's direct effects of Government decisions in EFO Table 3.1, 7.6 in 2026-27 and 3.5 in 2027-28 (uk_obr_tables). We infer from that match that IFS restates the OBR, so we give both rows medium confidence. + - Neither the total effect including indirect effects (6.3 and 7.0) nor Table 6.1's total of policy decisions (7,595 and 2,475 £m, uk_hmt_tables) matches both years. +- `ab2026__package_ifs_announced_tax_changes` (1 row, from Flourish 30726047): the tax change that Budget 2026 announced, at the forecast horizon, as a share of GDP as forecast at the event. + - In the chart's footer IFS credits its own calculations from the OBR's October 2026 EFO, policy measures database and historical official forecasts database. + - IFS lists no Table 6.1 line and names no OBR grouping, and one rounded figure can't tie the row to the scope of `ab2026__package_total_tax_policy_decisions` or of the OBR's tax decisions since March. So we propose a key in IFS's own terms. ## Keys the addendum changed (rules 3 and 9) | Rows (first lane, 0-based) | Was | Now | Why | |---|---|---|---| -| 45, 46, 52 (£5bn, £140, 35%) | proposed `ab2026__package_ifs_cost_of_living_support` | `ab2026__package_household_energy_vat_and_fuel_duty` | IFS's three measures are the energy payment, the whole electricity VAT cut and the fuel duty cut, exactly lines 1, 2 and 3 (README rule 3, corrected). Lines 1+2+3 print 5,040 over 2026-27 and 2027-28, IFS's £5bn. Since 4 November the £5bn row (0-based row 42) is `cumulative`, period 2028, `fiscal_year`, with the two years in `horizon`, as RF dates its two-year totals, and its note no longer says the scope differs from lines 1–3 (review A, K-7(e)). | +| 45, 46, 52 (£5bn, £140, 35%) | proposed `ab2026__package_ifs_cost_of_living_support` | `ab2026__package_household_energy_vat_and_fuel_duty` | IFS's three measures comprise the energy payment, the whole electricity VAT cut and the fuel duty cut, exactly lines 1, 2 and 3 (README rule 3, corrected). On lines 1+2+3 HM Treasury prints 5,040 over 2026-27 and 2027-28, IFS's £5bn. Since 4 November we mark the £5bn row (0-based row 42) `cumulative`, period 2028, `fiscal_year`, with the two years in `horizon`, as RF dates its two-year totals. We no longer say in its note that the scope differs from lines 1–3 (review A, K-7(e)). | | 48 (£36) | proposed `ab2026__package_ifs_electricity_vat_oct_2026_june_2027` | `ab2026__electricity_vat_zero_rate_to_july_2027` | Line 2 costs the whole cut, 1 October 2026 to 1 July 2027, including the weeks before the Budget. | | 89 (£1.8bn) | proposed `ab2026__package_ifs_self_assessment_payment_timing` | `ab2026__hmrc_tax_gap_package`, component | Table 6.3 row 3, "More timely payment" for Self Assessment, +1,815m in 2031–32: a part of line 45. | | 92 (£1bn) | proposed `ab2026__package_ifs_hmrc_compliance_staff_and_debt_collection` | `ab2026__hmrc_tax_gap_package`, component | Table 6.3 rows 7 (debt management, +575m) and 8 (compliance officers, +415m), 990 in 2031–32: a part of line 45. | -| 91 (£1.1bn) | proposed `ab2026__package_ifs_savings_and_property_income_tax` | proposed `ab2026__package_property_and_savings_income_rates` | The name README rule 2 gives exactly lines 61 and 62. | -| 55–57 | proposed `ab2026__package_ifs_gas_standing_charge_reallocation` | proposed `ab2026__gas_standing_charge_moved_to_unit_rates` | A measure with no Table 6.1 line is `ab2026__`; the RF family's spelling. | +| 91 (£1.1bn) | proposed `ab2026__package_ifs_savings_and_property_income_tax` | proposed `ab2026__package_property_and_savings_income_rates` | The name we give exactly lines 61 and 62 under README rule 2. | +| 55–57 | proposed `ab2026__package_ifs_gas_standing_charge_reallocation` | proposed `ab2026__gas_standing_charge_moved_to_unit_rates` | We give a measure with no Table 6.1 line `ab2026__`; the RF family's spelling. | | 61 (£1.2bn per year) | proposed `ab2026__package_ifs_defence_investment_plan_funding` | dropped | See below. | -**Defence Investment Plan funding.** The addendum asked whether IFS's "£1.2bn per year gap in Defence Investment Plan funding" is an estimate (keep, as context) or not. The first lane's own note calls it the government's DIP funding announcement, restated: the Budget "filled" the gap. That makes it a spending allocation, which rule 8 excludes, so it is dropped rather than kept as context. It is close to, but not exactly, Table 6.1 line 41 (capital reallocation to fully fund the DIP: +1,100, +1,135, +1,180 in 2029-30 to 2031-32), so it is not keyed there either. +**Defence Investment Plan funding.** The coordinator asked in the addendum whether IFS estimates the "£1.2bn per year gap in Defence Investment Plan funding" (keep, as context) or not. In its own note the first lane calls it the government's DIP funding announcement, restated: the Budget "filled" the gap. That makes it a spending allocation, which we exclude under rule 8, so we dropped it rather than keep it as context. It comes close to, but doesn't equal, Table 6.1 line 41 (capital reallocation to fully fund the DIP: +1,100, +1,135, +1,180 in 2029-30 to 2031-32), so we don't key it there either. ## Rows from addendum sections 1 and 2 ### The two pages (11 rows) -The rows cite each page's saved `
` text by line. +We cite each page's saved `
` text by line. The initial response (rows 87 to 91): -- Line 16: £13 billion and £15 billion, `fiscal_headroom` in 2029–30 before and after the Budget's policy decisions (`policy_scenario` pre-measures and post-measures). They restate the OBR, which the sentence names. The OBR family stages £12.5 billion and £14.5 billion against the target "Current budget to be in surplus by 2029-30" (uk_obr_tables); the rows keep the page's rounded figures. -- Line 18: £8 billion and £4 billion, `decisions_effect_on_borrowing` in 2026-27 and 2027-28 (this year and next at the 28 October Budget), under the proposed key above. We infer the OBR attribution from the rounded match to Table 3.1, so both rows have medium confidence. -- Line 28: 80%, `affected_share` 0.8. IFS's estimate of the share of households in England that receive the £100 payment, keyed to `ab2026__household_energy_payment_100_bands_a_to_d`; `reform_hint` keeps the £100. +- Line 16: £13 billion and £15 billion, `fiscal_headroom` in 2029–30 before and after the Budget's policy decisions (`policy_scenario` pre-measures and post-measures). IFS restates the OBR and names it in the sentence. In the OBR family we stage £12.5 billion and £14.5 billion against the target "Current budget to be in surplus by 2029-30" (uk_obr_tables); here we keep the page's rounded figures. +- Line 18: £8 billion and £4 billion, `decisions_effect_on_borrowing` in 2026-27 and 2027-28 (this year and next at the 28 October Budget), under `ab2026__package_ifs_new_giveaways` above. We infer the OBR attribution from the rounded match to Table 3.1, so we give both rows medium confidence. +- Line 28: 80%, `affected_share` 0.8: IFS's estimate of the share of households in England that receive the £100 payment. We key it to `ab2026__household_energy_payment_100_bands_a_to_d` and keep the £100 in `reform_hint`. -The opening remarks (rows 92 to 97; the saved text holds the whole speech on line 6): +The opening remarks (rows 92 to 97; we saved the whole speech on line 6 of the text): -- 98: proposed metric `measure_count` in proposed unit `measures`, `measure_key` null, no proposed key. A count of the Budget's measures, staged as the AB2025 and AB2026 uk_thinktank_misc families stage such counts (`own`, `different_model`). Table 6.1 numbers its lines 1 to 98 (`data/uk/ab2026_measures.json`). -- £17 billion: proposed metric `spending_pressure`, IFS's estimate of the top-up that avoiding real-terms cuts could need. The page prints its year as "2030-21", a typo, most likely for 2030–31. The row keeps "2030-21" in the quote and `conditions.horizon`, sets no `fy`, takes 2026 as a publication-year marker, and has medium confidence. -- £11 billion: `spending_pressure` for defence plans, on top of the £17 billion. The page prints no year; Boileau slide 4 prints £11 billion more defence spending in 2030–31 (row 55). Medium confidence and a 2026 marker. -- 25% and 40%: proposed metric `reserve_allocated_share`, the shares of the capital and day-to-day reserves already spent, restated `administrative_fact`. Neither the remarks nor Boileau slide 11, which prints the same two shares, names a source, so all four rows infer the restated attribution and have medium confidence. +- 98: proposed metric `measure_count` in proposed unit `measures`, `measure_key` null, no proposed key. Miller counts the Budget's measures, and we stage the count as we stage such counts in the AB2025 and AB2026 uk_thinktank_misc families (`own`, `different_model`). HM Treasury numbers Table 6.1's lines 1 to 98 (`data/uk/ab2026_measures.json`). +- £17 billion: proposed metric `spending_pressure`, IFS's estimate of the top-up that avoiding real-terms cuts could need. IFS prints its year as "2030-21", a typo, most likely for 2030–31. We keep "2030-21" in the quote and `conditions.horizon`, set no `fy`, use 2026 as a publication-year marker and give the row medium confidence. +- £11 billion: `spending_pressure` for defence plans, on top of the £17 billion. IFS prints no year on the page; on Boileau slide 4 it prints £11 billion more defence spending in 2030–31 (row 55). We give the row medium confidence and a 2026 marker. +- 25% and 40%: proposed metric `reserve_allocated_share`, the shares of the capital and day-to-day reserves already spent, restated `administrative_fact`. IFS names a source neither in the remarks nor on Boileau slide 11, where it prints the same two shares. So in all four rows we infer the restated attribution and give medium confidence. - 80%: `affected_share` 0.8 of households, for the £100 payment through council tax bands A-D, keyed to `ab2026__household_energy_payment_100_bands_a_to_d`. ### The seven visualisations (46 rows) -`.ab2026-primaries/uk_ifs_ab2026/browser/data-items.json` lists the seven data items in the collection's key charts, each with its Flourish id and lead sentence. The rows read each chart's data cells from its public embed (see "Access recipe and primary inventory"). Following the addendum, they stage the October 2026 forecast values, and the March 2026 values where a chart compares the two, as restated OBR figures unless IFS computed them; IFS's own calculations are `own`. Outturns stay unstaged unless a lead states one as its finding. +In `.ab2026-primaries/uk_ifs_ab2026/browser/data-items.json` we list the seven data items in the collection's key charts, each with its Flourish id and lead sentence. We read each chart's data cells from its public embed (see "Access recipe and primary inventory"). Following the addendum, we stage the October 2026 forecast values, and the March 2026 values where a chart compares the two. We mark them restated OBR figures unless IFS computed them, and mark IFS's own calculations `own`. We leave outturns unstaged unless IFS states one as the finding in a lead sentence. -| Flourish id | Rows | Metric and unit | What the rows hold | Attribution | +| Flourish id | Rows | Metric and unit | What we stage | Attribution | |---|--:|---|---|---| | 30734907 | 5 | `real_household_disposable_income_index` (proposed), `index` (proposed) | the 2020s forecast path, years 6 to 10 of the decade, read as 2026 to 2030; the lead's finding is the year-10 value | `own`: IFS builds the decade index from OBR and ONS data, and the addendum classes this chart as own | | 30726471 | 11 | `current_budget_deficit` (proposed), `gbp` | the October 2026 forecast for 2026-27 to 2031-32 and the March 2026 forecast for 2026-27 to 2030-31 | `restated` OBR | | 30726788 | 2 | `fiscal_headroom`, `share` | headroom at the March 2026 and October 2026 forecasts, as fractions of GDP | `restated` OBR | -| 30726917 | 11 | `real_gdp_index` (proposed), `index` (proposed) | the March 2026 forecast for 2026-27 to 2030-31 and the October 2026 forecast for 2026-27 to 2031-32, with 2025-26 = 100 | `restated` OBR; the addendum classes this chart as restated | +| 30726917 | 11 | `real_gdp_index` (proposed), `index` (proposed) | the March 2026 forecast for 2026-27 to 2030-31 and the October 2026 forecast for 2026-27 to 2031-32, with 2025-26 = 100 | `restated` OBR; we class this chart as restated under the addendum | | 30726939 | 1 | `real_income_growth`, `percent` | 0.4% a year over the 2024-2029 parliament | `own`: the footer credits IFS calculations | -| 30726047 | 1 | `revenue_change`, `percent` | Budget 2026's tax change at the forecast horizon, 0.2% of GDP, with the proposed key above | `own`: the footer credits IFS calculations | +| 30726047 | 1 | `revenue_change`, `percent` | Budget 2026's tax change at the forecast horizon, 0.2% of GDP, with `ab2026__package_ifs_announced_tax_changes` above | `own`: the footer credits IFS calculations | | 30727239 | 15 | `taxpayer_count`, `persons` | the three forecast series (the freeze at the March and October 2026 forecasts, and CPI uprating at the October forecast), 2026-27 to 2030-31 | `restated` OBR: the footer cites the OBR EFO, October 2026 | -Only the tax-change row carries a key. The other 45 hold forecast levels or IFS indicators, not the effect of one measure, so they have `measure_key` null and no proposed key. +We key only the tax-change row. In the other 45 IFS gives forecast levels or its own indicators, not the effect of one measure, so we give them `measure_key` null and no proposed key. ### Conventions for these rows -- **Quotes and locators.** Each Flourish row quotes "row, series: value", and its locator names the visualisation, the data row and the series. `value_raw` is the cell as the embed serves it, at full precision, whatever the chart displays. -- **Units.** The deficit chart's y-axis title is "£ billion", so its cells scale by 1e9 to GBP. The taxpayer chart's cells are thousands: the lead's 10.8 million is the 2030-31 October freeze cell, 10800, and the chart's number format sets a constant of 1000. They scale by 1,000 to persons. The headroom cells are fractions of GDP, staged as served in unit `share`; the chart's number format sets a constant of 100, one decimal place and a % suffix. Index levels keep proposed unit `index`, not the registered `index_0_1`. The parliament and tax-change cells are percentages: the number format adds a % suffix, and the tax column's name says "% of GDP". -- **Periods.** Fiscal-year rows take the end year and `conditions.fy`. The decade chart labels its rows "Year of decade" 0 to 10; the rows read years 6 to 10 of the 2020s as calendar 2026 to 2030, which the chart does not print, so they have medium confidence. The headroom and tax-change cells are fiscal-event snapshots (`point_in_time`, period 2026). The parliament average takes the parliament's end year, 2029, and keeps 2024-2029 in `conditions.horizon`. Text rows with no printed year take 2026 as a disclosed publication-year marker and import no year from the decks. -- **Vintages.** `conditions.forecast_vintage` follows the series or row label, March 2026 or October 2026. The decade chart's footer names the OBR's October 2026 EFO, and the parliament chart's lead names OBR forecasts, so those rows read October 2026. -- **Signs.** As served. A negative deficit cell is a surplus; the chart shades the area below zero "Current budget surplus". -- **Sources.** Restated OBR rows carry `source_model: obr_efo_forecast` and `benchmark_class: different_model`; IFS's own calculations carry `arithmetic`. The two reserve shares are `administrative_fact`, as in the Boileau rows. +- **Quotes and locators.** We quote each Flourish row as "row, series: value", and name the visualisation, the data row and the series in its locator. We take `value_raw` as the cell the embed serves, at full precision, whatever the chart displays. +- **Units.** + - IFS titles the deficit chart's y-axis "£ billion", so we scale its cells by 1e9 to GBP. + - IFS gives the taxpayer chart's cells in thousands: the lead's 10.8 million equals the 2030-31 October freeze cell, 10800, and in the chart's number format IFS sets a constant of 1000. We scale them by 1,000 to persons. + - IFS gives the headroom cells as fractions of GDP, which we stage as served in unit `share`; in the chart's number format IFS sets a constant of 100, one decimal place and a % suffix. + - We keep index levels in the proposed unit `index`, not the registered `index_0_1`. + - IFS gives the parliament and tax-change cells as percentages: it adds a % suffix in the number format and writes "% of GDP" in the tax column's name. +- **Periods.** + - We give fiscal-year rows the end year and `conditions.fy`. + - IFS labels the decade chart's rows "Year of decade" 0 to 10. We read years 6 to 10 of the 2020s as calendar 2026 to 2030, which IFS doesn't print, so we give those rows medium confidence. + - We stage the headroom and tax-change cells as fiscal-event snapshots (`point_in_time`, period 2026). + - We put the parliament average at the parliament's end year, 2029, and keep 2024-2029 in `conditions.horizon`. + - For text rows with no printed year we use 2026 as a disclosed publication-year marker, and import no year from the decks. +- **Vintages.** We set `conditions.forecast_vintage` from the series or row label, March 2026 or October 2026. IFS names the OBR's October 2026 EFO in the decade chart's footer, and OBR forecasts in the parliament chart's lead, so we set those rows to October 2026. +- **Signs.** We stage cells as served. We read a negative deficit cell as a surplus; IFS shades the area below zero "Current budget surplus". +- **Sources.** We give restated OBR rows `source_model: obr_efo_forecast` and `benchmark_class: different_model`, and IFS's own calculations `arithmetic`. We mark the two reserve shares `administrative_fact`, as in the Boileau rows. ## Not staged (rules 8 and 9) ### Deck rows the addendum dropped -The addendum dropped 19 of the first lane's 106 rows, all under rule 8: +Under the addendum we dropped 19 of the first lane's 106 rows, all under rule 8: | Category | Rows | Examples (deck, slide) | |---|--:|---| @@ -131,24 +150,31 @@ The addendum dropped 19 of the first lane's 106 rows, all under rule 8: | Government targets | 3 | defence at 3% of GDP by 2030 (Boileau 4, repeated on 8); the DIP's earlier 2.7% in 2029–30 (Boileau 4) | | Inputs of IFS's own spending scenarios | 2 | defence held at 2.7% of GDP, or at 3% by 2030–31 (Boileau 7 and 9); the scenarios' computed growth rates stay | | Financing and allocations | 2 | ~£1bn a year of NWF and BBB financing for defence (Boileau 5); the £1.2bn a year of DIP funding (Boileau 4) | -| The base a measure moves | 2 | the £108 average gas standing charge and its 13% share of a typical gas bill (Wernham 8), as the red book dropped its £108 | +| The base a measure moves | 2 | the £108 average gas standing charge and its 13% share of a typical gas bill (Wernham 8), as we dropped the red book's £108 | -Each dropped parameter that names a measure stays in the kept rows' `reform_hint`. The 25% capital and 40% day-to-day reserve shares (Boileau 11) stay: the addendum lists them as estimates. The bond spread and debt-ownership context rows (Stockton 7) are not design and stay as context. +When a dropped parameter names a measure, such as the £100 payment, we keep it in the kept rows' `reform_hint`. We keep the 25% capital and 40% day-to-day reserve shares (Boileau 11), because the coordinator lists them as estimates in the addendum. We keep the bond spread and debt-ownership context rows (Stockton 7) as context, since they aren't design. ### Figures on the two pages (11) -| Page | Figure | Times printed | Why the rows leave it out | +| Page | Figure | Times printed | Why we leave it out of the rows | |---|---|--:|---| | Initial response | 3% of GDP on defence by 2030 | 2 | a government target (rule 8) | -| Initial response | the £100 payment | 1 | a measure parameter (rule 8); row 91's `reform_hint` keeps it | +| Initial response | the £100 payment | 1 | a measure parameter (rule 8); we keep it in row 91's `reform_hint` | | Opening remarks | 3% of GDP on defence | 3 | a government target | | Opening remarks | 3.5% by 2035 | 1 | a government target | | Opening remarks | 2.7% of GDP from next year to 2029–30 | 1 | the government's defence spending plan, a target | -| Opening remarks | the £100 payment | 1 | a measure parameter; row 97's `reform_hint` keeps it | +| Opening remarks | the £100 payment | 1 | a measure parameter; we keep it in row 97's `reform_hint` | | Opening remarks | property values 35 years out of date | 1 | it describes council tax valuations and estimates no Budget effect | | Opening remarks | local areas to retain 100% of business rates | 1 | the design of a devolution plan (rule 8) | -The pages' other numbers are not figures to stage: publication dates, titles and target dates (2030, 2035); horizons ("next five years", and "next three years", which row 95 keeps in `conditions.horizon`); the fiscal rule's year, 2029–30, which the headroom rows carry as `fy`; "2030-21", which row 93 keeps; the name of the "10 year plan"; the G7; the count of "two big problems"; and "one-off". The remarks' "almost halved" prints no number, and no row turns it into one. +We don't count the pages' other numbers as figures to stage: +- publication dates, titles and target dates (2030, 2035); +- horizons ("next five years", and "next three years", which we keep in row 95's `conditions.horizon`); +- the fiscal rule's year, 2029–30, which we record as `fy` on the headroom rows; +- "2030-21", which we keep in row 93; +- the name of the "10 year plan"; the G7; the count of "two big problems"; and "one-off". + +In her remarks Miller gives no number with "almost halved", and we turn it into none. ### Data cells in the seven visualisations (256 of 302) @@ -163,56 +189,57 @@ The pages' other numbers are not figures to stage: publication dates, titles and | 30727239 | 58 | 15 | 43 | the Current plans series, 1990-91 to 2025-26 (36); the two freeze series' 2025-26 cells (2), each 8510, the Current plans value for that year; and the CPI-uprating series for 2021-22 to 2025-26 (5), a counterfactual path for the years before the forecast years, whose 2021-22 cell equals Current plans' 4950 | | **Total** | **302** | **46** | **256** | | -Outside the data cells, the headroom chart draws a reference line labelled with the 2010s average headroom, 0.9% of GDP: a historical average, which no row stages. The lead sentences in `data-items.json` belong to no manifested primary. Each estimate a lead states repeats a staged cell (0.4% a year is row 127, around 0.2% of GDP is row 128, 10.8 million is row 142), and their other numbers are dates. Axis titles define index bases, and apart from the labelled 0.9% line, no tick or guide adds an estimate. No row reads a value off an image, a bar, a line or a tick. +Outside the data cells, IFS draws a reference line on the headroom chart labelled with the 2010s average headroom, 0.9% of GDP: a historical average, which we don't stage. The lead sentences in `data-items.json` belong to no manifested primary. Where IFS states an estimate in a lead, it repeats a staged cell (0.4% a year in row 127, around 0.2% of GDP in row 128, 10.8 million in row 142), and it uses other numbers only for dates. IFS defines index bases in the axis titles, and apart from the labelled 0.9% line, adds no estimate in any tick or guide. We read no value off an image, a bar, a line or a tick. ## Unreached -We couldn't fetch two of the sources the addendum names, and two others held nothing to stage: +We couldn't fetch two of the sources the coordinator named in the addendum, and we found nothing to stage in two others: -- **The seven IFS data-item pages.** ifs.org.uk refuses curl, and the addendum rules out any other route, so each manifest line names its data item's URL and hashes the chart's public Flourish embed instead. The rows do not use the "Download data" link that each chart's footer template names. -- **The explainer and the fiscal devolution comment.** The browser read of 29 October found only an embedded video in the first and only design parameters in the second, so neither has a manifest line or rows. -- **The briefing event page** (https://ifs.org.uk/events/autumn-budget-2026-ifs-analysis) refused the first lane's curl with a Cloudflare challenge (see below); no primary holds it. +- **The seven IFS data-item pages.** ifs.org.uk refuses curl, and the coordinator rules out any other route in the addendum. So in each manifest line we name the data item's URL and hash the chart's public Flourish embed instead. We don't use the "Download data" link that IFS names in each chart's footer template. +- **The explainer and the fiscal devolution comment.** In the browser read of 29 October we found only an embedded video in the first and only design parameters in the second, so we gave neither a manifest line or rows. +- **The briefing event page** (https://ifs.org.uk/events/autumn-budget-2026-ifs-analysis). IFS's server refused the first lane's curl with a Cloudflare challenge (see below), and we hold it in no primary. ## Verification (3 November 2026) -- **Checker:** `PYTHONPATH=. python pipeline/ab2026_harvest_contract.py sources/harvest-uk-ab2026-2026-10-29/uk_ifs_ab2026` reports 144 rows, 0 problems. -- **Determinism:** two runs of `stage.py` write the same gzip, sha256 `40331bc6beb36fc78d2e10731730c4c8a15942d4419bd840d0cebc95988e6125`. The 87-row file before sections 1 and 2 had `2e8b2dc5a85bb66fb11431f5c20565a40b2cf546e7edaabedd3a6c56ddb68d49`. COUNTS.json pins the new hash. -- **The 87 deck rows:** the new gzip's first 87 serialised lines match the 87-row file byte for byte (3,247 leaf fields, 0 differences), and manifest lines 1 to 5 match theirs. -- **Every new row:** a script re-opened all 57 rows at their locators: the saved text's line, or the embed's data row and series, decoding the embed afresh with every number kept as served text. It checked the primary's sha256 against the row and the manifest, the quote against the line or cell, `value_raw` against the cell, and the value, sign, unit, period and forecast vintage. All 57 pass. Nine planted errors (value, sign, period, `fy`, series, unit, vintage, quote and hash) each made it fail. -- **Cell accounting:** the same decoder lists every non-null cell in the seven embeds: 302, of which the rows stage 46. -- **Flourish bytes:** fetches on 29 October and 3 November returned byte-identical embeds. -- **Tests:** `PYTHONPATH=. python -m pytest -n 4 --noconftest tests/test_ab2026_harvest_contract.py tests/test_uk_ab2026_harvest.py`: 52 passed. `--noconftest` skips the conftest's build of the derived database, which neither file uses. -- **Generated files:** `pipeline/ab2026_proposed_keys.py` and `pipeline/harvest_counts.py --pinned-on 2026-10-29` regenerate PROPOSED_KEYS.md and COUNTS.json, and both `--check` runs pass. -- **After the review:** commit f4c7c94a's fixes and README rule 10's move of row 76 changed no value, sign or year. COUNTS.json pins the gzip's current sha256, and `stage.py` still writes it byte for byte. +- **Checker.** `PYTHONPATH=. python pipeline/ab2026_harvest_contract.py sources/harvest-uk-ab2026-2026-10-29/uk_ifs_ab2026` reports 144 rows, 0 problems. +- **Determinism.** Two runs of `stage.py` write the same gzip (sha256 `40331bc6beb36fc78d2e10731730c4c8a15942d4419bd840d0cebc95988e6125`), which COUNTS.json pins; the 87-row file before sections 1 and 2 hashed to `2e8b2dc5a85bb66fb11431f5c20565a40b2cf546e7edaabedd3a6c56ddb68d49`. +- **The 87 deck rows.** The new gzip's first 87 serialised lines match the 87-row file byte for byte (3,247 leaf fields, 0 differences), and manifest lines 1 to 5 match theirs. +- **Every new row.** We re-opened all 57 rows with a script at their locators: the saved text's line, or the embed's data row and series, decoding the embed afresh with every number kept as served text. The script checked the primary's sha256 against the row and the manifest, the quote against the line or cell, `value_raw` against the cell, and the value, sign, unit, period and forecast vintage. All 57 pass. We planted nine errors (value, sign, period, `fy`, series, unit, vintage, quote and hash), and each made it fail. +- **Cell accounting.** The same decoder lists every non-null cell in the seven embeds: 302, of which we stage 46. +- **Flourish bytes.** Our fetches on 29 October and 3 November returned byte-identical embeds. +- **Tests.** `PYTHONPATH=. python -m pytest -n 4 --noconftest tests/test_ab2026_harvest_contract.py tests/test_uk_ab2026_harvest.py`: 52 passed. `--noconftest` skips the conftest's build of the derived database, which neither file uses. +- **Generated files.** `pipeline/ab2026_proposed_keys.py` and `pipeline/harvest_counts.py --pinned-on 2026-10-29` regenerate PROPOSED_KEYS.md and COUNTS.json, and both `--check` runs pass. +- **After the review.** Commit f4c7c94a's fixes and README rule 10's move of row 76 changed no value, sign or year. COUNTS.json pins the gzip's current sha256, and `stage.py` still writes it byte for byte. -The scripts and their outputs sit in the ignored `.ab2026-primaries/uk_ifs_ab2026/a3/`. +We keep the scripts and their outputs in the ignored `.ab2026-primaries/uk_ifs_ab2026/a3/`. ## Access recipe and primary inventory ### Decks (first lane, 29 October) -Fresh fetches began **2026-10-29 15:43:43 UTC**, using exactly: +The first lane began fresh fetches at **2026-10-29 15:43:43 UTC**, with exactly: ```sh curl -sSL -A "Mozilla/5.0 (Macintosh; Intel Mac OS X 14_0) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/17.0 Safari/605.1.15" URL ``` -The fetch also saved response headers and HTTP status and imposed a 50-second -timeout. Bytes are under `.ab2026-primaries/uk_ifs_ab2026/`, outside the vendored -family. SHA256s were checked with `shasum -a 256`. PDFs were read with -`pdftotext -layout` and rendered for visual checks of labels and embedded images. -None of the PDF annotations links to a chart-data download. +The lane also saved response headers and HTTP status, and set a 50-second +timeout. It saved the bytes under `.ab2026-primaries/uk_ifs_ab2026/`, outside the +vendored family, and checked SHA256s with `shasum -a 256`. It read the PDFs with +`pdftotext -layout` and rendered them for visual checks of labels and embedded +images. None of IFS's PDF annotations links to a chart-data download. -Stockton's fresh file is byte-for-byte identical to assigned `stockton.pdf`, -confirming it is **Weathering-the-fiscal-storm-IS.pdf**. The other three current -decks also match their supplied local copies. TW_0 returned HTTP 403; the -supplied local PDF was copied to `TW_0-local.pdf` and hashed. All rows from the -cost-of-living presentation cite TW_1, not the superseded version. Full TW_0/TW_1 -changes are documented below; no numerical claim changed. +Stockton's fresh file matches the assigned `stockton.pdf` byte for byte, +which confirms it as **Weathering-the-fiscal-storm-IS.pdf**. The other three +current decks also match their supplied local copies. IFS's server returned HTTP +403 for TW_0, so the lane copied the supplied local PDF to `TW_0-local.pdf` and +hashed it. All rows from the cost-of-living presentation cite TW_1, not the +superseded version. We document the full TW_0/TW_1 changes below; IFS changed no +numerical claim. ### Browser text (the initial response and the opening remarks) -On 29 October, between 16:56Z and 16:58Z, the harvest read the two pages in the Claude desktop app's built-in browser, as an ordinary page visit with no challenge solved. It fetched each page's HTML in-page with `fetch(location.href)` and hashed it. Two fetches of the initial response gave the same length, 257,037 bytes, with different hashes: the HTML embeds per-request tokens. So the primary is the text of each page's `
` element, and each manifest note records the HTML's byte count and one fetch's hash. Both pages print publication on 28 October 2026; the remarks are Helen Miller's opening remarks at the 29 October briefing. +On 29 October, between 16:56Z and 16:58Z, we read the two pages in the Claude desktop app's built-in browser, as an ordinary page visit with no challenge solved. We fetched each page's HTML in-page with `fetch(location.href)` and hashed it. Two fetches of the initial response gave the same length, 257,037 bytes, with different hashes, because IFS embeds per-request tokens in the HTML. So we take the text of each page's `
` element as the primary, and record the HTML's byte count and one fetch's hash in each manifest note. IFS dates both pages 28 October 2026; on one it gives Helen Miller's opening remarks at the 29 October briefing. ### Flourish public data (the seven visualisations) @@ -222,7 +249,10 @@ curl -sSL --max-time 50 -A "$UA" https://public.flourish.studio/visualisation/ID curl -sSL --max-time 50 -A "$UA" https://flo.uri.sh/visualisation/ID/embed # the hashed primary ``` -Both answered HTTP 200 for all seven ids on 3 November. Each embed is an HTML page whose inline script assigns `_Flourish_data`, `_Flourish_data_column_names` and `_Flourish_settings` as JSON; a JSON decoder reads them from the assignment onward. `_Flourish_data_timestamps` holds a JavaScript `new Date(...)`, which needs its own parse. Each manifest note gives the data item's URL, the public page's bytes and hash, the embed URL, the fetch time, the Flourish version, the data's last update and the lead sentence from `data-items.json`. +Both answered HTTP 200 for all seven ids on 3 November. +- Flourish serves each embed as an HTML page whose inline script assigns `_Flourish_data`, `_Flourish_data_column_names` and `_Flourish_settings` as JSON, and we read them with a JSON decoder from the assignment onward. +- Flourish stores `_Flourish_data_timestamps` as a JavaScript `new Date(...)`, which we parse separately. +- In each manifest note we give the data item's URL, the public page's bytes and hash, the embed URL, the fetch time, the Flourish version, the data's last update and the lead sentence from `data-items.json`. ### Inventory @@ -245,25 +275,25 @@ Both answered HTTP 200 for all seven ids on 3 November. Each embed is an HTML pa ### HTML pages that refused curl -The first lane's record of the HTML pages that returned a Cloudflare challenge (HTTP 403) to its curl at 15:43:43 UTC; "Browser text" above and "Unreached" give what has become of each: +When the first lane ran curl at 15:43:43 UTC, these HTML pages answered with a Cloudflare challenge (HTTP 403). "Browser text" above and "Unreached" say what became of each: - https://ifs.org.uk/articles/autumn-budget-2026-initial-response - https://ifs.org.uk/articles/autumn-budget-2026-explained - https://ifs.org.uk/collections/autumn-budget-2026 - https://ifs.org.uk/events/autumn-budget-2026-ifs-analysis -No assigned local HTML copies exist. No attempt was made to bypass the blocks. -Challenge pages are retained only as fetch evidence in the ignored primaries -directory; they are not IFS publications and are not manifested. No rows are -staged from them. The explainer's existence/content cannot be confirmed from its -challenge response. The blocked collection and event pages prevent discovering -additional AB2026 items or linked xlsx/csv downloads. No data file was identified -inside the reachable PDFs; this is an access limit, not a finding that IFS posted -no data. No Wayback, alternative domain or other bypass was attempted. +Nobody assigned local HTML copies, and the lane made no attempt to bypass the +blocks. We keep the challenge pages only as fetch evidence in the ignored +primaries directory; IFS didn't publish them, and we don't manifest them or stage +rows from them. We can't confirm the explainer's existence or content from its +challenge response. We couldn't discover further AB2026 items or linked +xlsx/csv downloads on the blocked collection and event pages. We found no data +file inside the reachable PDFs; that reflects an access limit, not a finding that +IFS posted no data. We attempted no Wayback, alternative domain or other bypass. ## Proposed metrics and units -The table lists every proposed metric in the 144 rows. +We list every proposed metric in the 144 rows. | Proposed metric | Rows | Unit(s) | Where and what | |---|--:|---|---| @@ -292,271 +322,337 @@ The table lists every proposed metric in the 144 rows. | `taxpayer_gain` | 1 | gbp | Adam slide 3: the maximum Founders' Relief gain, £1.4m | | `ten_year_bond_spread_other_g7` | 1 | percentage_points | Stockton slide 7: 134bp | -The proposed units are `index` (16 rows, the two index charts) and `measures` (the measure count); no row proposes a baseline. The first lane's rows proposed `affected_share` and `cpi_inflation_rate`, which the Metric registry carries; since 4 November (review A's H-6(a)) those four rows (Wernham slides 2 and 5) carry them in `metric`, and the new rows use the registered `affected_share` too. The new rows also use the registered metrics `taxpayer_count` (15), `fiscal_headroom` (4), `decisions_effect_on_borrowing` (2), `real_income_growth` (1) and `revenue_change` (1). +We propose the units `index` (16 rows, the two index charts) and `measures` (the measure count), and propose no baseline on any row. The first lane proposed the registered metrics `affected_share` and `cpi_inflation_rate`. Since 4 November (review A's H-6(a)) we record them in `metric` on those four rows (Wernham slides 2 and 5), and use the registered `affected_share` in the new rows too. In the new rows we also use the registered metrics `taxpayer_count` (15), `fiscal_headroom` (4), `decisions_effect_on_borrowing` (2), `real_income_growth` (1) and `revenue_change` (1). ## Family conventions -- Fiscal years use their end year and `conditions.fy`. Calendar years, - cross-fiscal-year nine-month support, and publication-date snapshots have - no `fy`; full horizons and missing dates are disclosed. Unspecified-year - Founders' Relief annual figures use annual 2026 publication markers and +- We give fiscal years their end year and `conditions.fy`. We give calendar + years, cross-fiscal-year nine-month support and publication-date snapshots + no `fy`, and disclose full horizons and missing dates. For Founders' Relief + annual figures with no year, we use annual 2026 publication markers and medium confidence, not an inferred implementation year. -- Monetary magnitudes and signs are verbatim. £bn and £m are normalised to - GBP. Percentage forecasts and rates stay percent; recipient shares divide - by 100. Fiscal cells say percent of GDP or revenues as printed, rather - than inventing a real-GDP unit. 134 basis points is 1.34 percentage points. - The fuel-duty parameter keeps proposed unit `pence_per_litre`. -- Real/cash and price bases stay as published. Nine-month household support - is not annualised. A printed positive magnitude followed by “decrease” - remains positive, with the direction recorded in conditions. -- All figures are `point`: no producer scenario is substituted for a central - OBR estimate and no confidence band is invented. “Central” would mean a - producer-labelled central forecast; the staged figures contain no such label. -- `arithmetic` is used for IFS calculations where the deck does not name - TAXBEN. Official costings/forecasts are `restated`, with original source - named in notes where supplied or inferable; administrative policy parameters - are distinguished from IFS estimates. IFS does not cite PolicyEngine here. -- Current-law/default baseline is null. No new baseline world is inferred. - Every condition value is a string. Quotes are under 15 words, contain the - exact value_raw, and refer to printed text, a data label or a numeric cell. +- We copy monetary magnitudes and signs verbatim and normalise £bn and £m to + GBP. We keep percentage forecasts and rates in percent, and divide recipient + shares by 100. We keep fiscal cells in percent of GDP or revenues as IFS + prints them, rather than invent a real-GDP unit. We convert 134 basis points + to 1.34 percentage points. We keep the proposed unit `pence_per_litre` for + the fuel-duty parameter. +- We keep real/cash and price bases as IFS publishes them, and don't + annualise nine-month household support. Where IFS prints a positive + magnitude followed by “decrease”, we keep it positive and record the + direction in conditions. +- We mark all figures `point`: we substitute no producer scenario for a + central OBR estimate and invent no confidence band. We would use “Central” + only for a forecast the producer labels central, and no staged figure has + such a label. +- We use `arithmetic` for IFS calculations where the deck doesn't name + TAXBEN. We mark official costings and forecasts `restated`, and name the + original source in notes where IFS supplies it or we can infer it. We + distinguish administrative policy parameters from IFS estimates. IFS doesn't + cite PolicyEngine here. +- We leave the current-law/default baseline null and infer no new baseline + world. We write every condition value as a string. We keep quotes under 15 + words; each contains the exact value_raw and comes from printed text, a data + label or a numeric cell. ## Stockton fiscal deck -39 rows: 38 printed numeric cells in slide 7's **Fiscal traffic lights** image -table, plus slide 8's “roughly 1 in 4 revisions” risk annotation. Text extraction -misses the entire image table; all cells were transcribed from the rendered -slide and independently audited. The table spans FY2026-27, 2027-28, 2029-30 and -2031-32; missing intermediate years were not invented. Net worth is explicitly -inverted and the signs were not reversed. The debt-interest denominator is -revenues, while debt ownership is a share of total debt. - -The table prints no source citation. OBR attribution for the 36 forecast cells -is inferred from the surrounding OBR forecast slides, not presented as an -explicit citation; these rows have medium confidence and disclose the inference. -Debt-ownership's administrative origin likewise has no named original provider -or observation date. The 134bp spread is an IFS market comparison with other -G7 economies whose source/weighting/date are unspecified. The latter two use a -2026 in-year snapshot. The 1-in-4 risk uses historical forecast revisions since -2010 excluding 2020-21; “Spring” is interpreted as Spring 2027 and disclosed. -All fiscal rows have null measure_key and no proposed measure. - -Not staged: slides 3–6 forecast/borrowing bars, diamonds and lines, and slide 8's -forecast/outturn lines, because they have no numeric data labels. Axis ticks, -chart origins and dates are not outcome rows. Slides 1–6 were visually checked -for hidden printed numbers. No headroom or tax/GDP number is printed there. +The first lane staged 39 rows: 38 printed numeric cells in slide 7's **Fiscal +traffic lights** image table, plus slide 8's “roughly 1 in 4 revisions” risk +annotation. Text extraction misses the whole image table, so the lane +transcribed every cell from the rendered slide and had it independently +audited. IFS's table spans FY2026-27, 2027-28, 2029-30 and 2031-32, and we +invented no missing intermediate years. IFS explicitly inverts net worth, and +we didn't reverse the signs. IFS divides debt interest by revenues, and gives +debt ownership as a share of total debt. + +IFS cites no source in the table. We infer OBR attribution for the 36 forecast +cells from the surrounding OBR forecast slides rather than from an explicit +citation, so we give these rows medium confidence and disclose the inference. +IFS likewise names no original provider or observation date for debt +ownership. IFS gives the 134bp spread as a market comparison with other G7 +economies, with unspecified source, weighting and date. We give these last two +a 2026 in-year snapshot. For the 1-in-4 risk IFS uses historical forecast +revisions since 2010, excluding 2020-21; we interpret “Spring” as Spring 2027 +and disclose it. We give all fiscal rows a null measure_key and no proposed +measure. + +The first lane didn't stage slides 3–6's forecast/borrowing bars, diamonds and +lines, or slide 8's forecast/outturn lines, because IFS prints no numeric data +labels on them. It staged no axis ticks, chart origins or dates as outcome rows, +and when it checked slides 1–6 by eye, it found no printed headroom or tax/GDP +number. ## Wernham cost-of-living deck -Tom Wernham component: 21 rows from the revised 17-slide Cost of living measures deck, cover dated 29 October 2026. Source slug `ifs`; own household-price calculations use `arithmetic` (the deck does not name TAXBEN), OBR forecasts use `obr_efo_forecast`, EDF's energy-price forecast uses `edf_energy_price_forecast`, and announced statutory parameters use `administrative_data` with restated attribution and administrative_fact benchmark. No chart bars were read for values. - -Access/version evidence: - -- TW_1 freshly fetched directly at 2026-10-29 15:43:43Z using the assigned Safari user agent. SHA256 `6ed174b22b4d22a34c8bbd450194fb2b4181a9d566b137913bee52ca49a0299b`, 222191 bytes. This hash is identical to the coordinator's local copy of the TW_1 PDF. -- TW_0 fresh URL returned HTTP403, so the assigned local file was copied to `.ab2026-primaries/uk_ifs_ab2026/TW_0-local.pdf`, hashed, and extracted using `pdftotext -layout`. SHA256 `abb63d9c4ecd21c3c9173077c5e92d8015c7cc30aebd12be159185d7d157f12a`, 231180 bytes; manifest access must be `local copy`. No bypass was attempted. -- Text diff: TW_0 has18 pages, TW_1 has17. TW_0 p4 already includes the80%/91% eligibility annotations; TW_1 p4 omits these. TW_0 p5 has those eligibility annotations overlapping the VAT/fuel commentary. TW_1 p5 relocates eligibility to the lower chart and VAT/fuel commentary to the upper chart. TW_0 p6 is another copy of the unannotated support chart and was removed. TW_0 p7–18 correspond to TW_1 p6–17. No printed numerical claim changes. All staged rows cite TW_1; TW_0 contributes zero rows but is manifested as the changed primary required by rule7. - -Coverage by measure_key: null14; `ab2026__household_energy_payment_100_bands_a_to_d`4; `ab2026__fuel_duty_5p_cut_extended_and_rpi_rise_deferred`2; `ab2026__bus_fare_cap_2_england_outside_london_2027`1. Within null: 6 unscored context rows; the remaining8 use proposed measure keys below. Rows per primary: TW_1=21, TW_0=0. - -Proposed measure scopes: the first lane's three Wernham proposals were re-keyed on 3 November (see "Keys the addendum changed" above). The cost-of-living package is exactly lines 1, 2 and 3, because line 2 costs the whole VAT cut including the weeks before the Budget; the full-period VAT row is line 2; the gas standing-charge rows use `ab2026__gas_standing_charge_moved_to_unit_rates`. - -Proposed metrics and units: - -- `cpi_inflation_rate`: printed OBR3.8% and1.9% CPI forecast levels for Q12027, not inflation effects from measures. -- `average_energy_bill_change`: existing AB2025 proposal; EDF£470 annual energy-price increase from January2027. -- `benefit_payment_amount`: £100 per eligible household one-off payment, rather than the average gain across all households. -- `fuel_duty_rate_cut`, proposed_unit `pence_per_litre`: printed5p extension; no negative sign is inserted. Per-litre basis follows the named policy, though the slide prints only5p. -- `statutory_tax_rate`:0% electricity VAT during October2026–June2027, not an effective tax rate. -- `average_household_support`: nine-month cash support, including indirect tax reductions; no conversion to annual gains or household income changes. Printed£140 headline and £24/£36/£80 all-household chart labels. -- `affected_share`: existing AB2025 proposal, payment receipt by80% of all England households and91% of the poorest30%; normalized to0.8 and0.91 shares. Household ranking is by spending, not income decile. -- `share_of_energy_price_increase_offset`: the printed "around35%" of forecast price increases covered by the three measures; normalized to0.35 share, no ratio derived from bars. -- `energy_standing_charge`: current annual gas standing charge£108; context. -- `energy_standing_charge_share`: current13% of a typical gas bill; normalized to0.13 share, context. -- `fare_cap`: £2 single bus fare cap, a restated policy parameter. Exact2027 timing and England-outside-London scope follow registered Table6.1 line4; the slide itself gives only "£2 cap on bus fares", and the row explicitly says this. - -Conventions: GBP amounts stay as printed in cash or2026-27 prices as specified; the £5bn aggregate is multiplied by1e9. Shares divide printed percentages by100, while CPI percentages and the printed1% income decrease remain percentages. Positive cost is cost to the Exchequer; positive household support is a gain. The printed1% decrease in real household disposable income is not re-signed; its direction is explicit in sign_convention. October2026–June2027 figures use annual time_basis, end-period marker2027 and horizon strings, with no fiscal-year condition because the nine months cross fiscal years. The1% RHDI decrease spansFY2024-25 toFY2026-27 and usesFY2026-27/end2027. No row is described as central: all figures are printed point estimates/parameters. For gas reform, year2026 is explicitly only a publication marker because no implementation year is printed. Baseline remains null; unsupported counterfactual definitions were not invented. +The first lane staged 21 Tom Wernham rows (15 after the 3 November addendum) from the revised 17-slide Cost of living measures deck, whose cover IFS dates 29 October 2026. We use source slug `ifs`. We use `arithmetic` for IFS's own household-price calculations (the deck doesn't name TAXBEN), `obr_efo_forecast` for OBR forecasts and `edf_energy_price_forecast` for EDF's energy-price forecast. The first lane gave announced statutory parameters `administrative_data`, restated attribution and an administrative_fact benchmark. The lane read no chart bars for values. + +Access and version evidence: + +- The lane freshly fetched TW_1 directly at 2026-10-29 15:43:43Z with the assigned Safari user agent: SHA256 `6ed174b22b4d22a34c8bbd450194fb2b4181a9d566b137913bee52ca49a0299b`, 222191 bytes. That hash matches the coordinator's local copy of the TW_1 PDF. +- IFS's server returned HTTP 403 for the fresh TW_0 URL. So the lane copied the assigned local file to `.ab2026-primaries/uk_ifs_ab2026/TW_0-local.pdf`, hashed it and extracted it with `pdftotext -layout`: SHA256 `abb63d9c4ecd21c3c9173077c5e92d8015c7cc30aebd12be159185d7d157f12a`, 231180 bytes. Its manifest access must read `local copy`. The lane attempted no bypass. +- Text diff: + - IFS prints 18 pages in TW_0 and 17 in TW_1. + - On TW_0 p4 IFS already includes the 80%/91% eligibility annotations, which TW_1 p4 omits. + - On TW_0 p5 the eligibility annotations overlap the VAT/fuel commentary. On TW_1 p5 IFS moves eligibility to the lower chart and the VAT/fuel commentary to the upper chart. + - IFS removed TW_0 p6, another copy of the unannotated support chart. TW_0 p7–18 correspond to TW_1 p6–17. + - IFS changed no printed numerical claim. In all staged rows we cite TW_1. We stage zero rows from TW_0, but we give it a manifest line as a changed primary, as README rule 7 requires. + +The first lane gave 14 of its rows a null measure_key, 4 `ab2026__household_energy_payment_100_bands_a_to_d`, 2 `ab2026__fuel_duty_5p_cut_extended_and_rpi_rise_deferred` and 1 `ab2026__bus_fare_cap_2_england_outside_london_2027`. Of the null rows, it staged 6 as unscored context and gave 8 its proposed measure keys. It staged all 21 rows from TW_1 (TW_1=21, TW_0=0). + +On 3 November we re-keyed the first lane's three Wernham proposals (see "Keys the addendum changed" above). The cost-of-living package covers exactly lines 1, 2 and 3, because line 2 costs the whole VAT cut, including the weeks before the Budget. We key the full-period VAT row to line 2, and the gas standing-charge rows to `ab2026__gas_standing_charge_moved_to_unit_rates`. + +The first lane proposed these metrics and units. Under the 3 November addendum we dropped the rows on `benefit_payment_amount`, `fuel_duty_rate_cut`, `statutory_tax_rate`, `energy_standing_charge`, `energy_standing_charge_share` and `fare_cap` (rule 8). + +- `cpi_inflation_rate`: IFS prints the OBR's 3.8% and 1.9% CPI forecast levels for Q1 2027, not inflation effects from measures. +- `average_energy_bill_change`: an existing AB2025 proposal; EDF's £470 annual energy-price increase from January 2027. +- `benefit_payment_amount`: the first lane proposed it for the £100 one-off payment per eligible household, not the average gain across all households. +- `fuel_duty_rate_cut`, proposed_unit `pence_per_litre`: IFS prints a 5p extension, and the first lane inserted no negative sign. It followed the named policy's per-litre basis, though IFS prints only 5p on the slide. +- `statutory_tax_rate`: the first lane proposed it for 0% electricity VAT during October 2026–June 2027, not an effective tax rate. +- `average_household_support`: nine-month cash support, including indirect tax reductions; we make no conversion to annual gains or household income changes. IFS prints the £140 headline and £24/£36/£80 all-household chart labels. At ingest we tally the £140, £24 and £36 as partial-year support. We score the £80 on `average_household_income_change`, because IFS averages January 2027's one-off Household Energy Payment across households in England and the whole payment falls in 2026-27. +- `affected_share`: an existing AB2025 proposal; payment receipt by 80% of all England households and 91% of the poorest 30%, normalised to 0.8 and 0.91 shares. IFS ranks households by spending, not income decile. +- `share_of_energy_price_increase_offset`: IFS's printed "around 35%" of forecast price increases covered by the three measures, normalised to a 0.35 share; we derive no ratio from bars. +- `energy_standing_charge`: the first lane staged the current annual gas standing charge of £108 on it as context. +- `energy_standing_charge_share`: the first lane staged the current 13% of a typical gas bill on it as context, normalised to a 0.13 share. +- `fare_cap`: the £2 single bus fare cap, a restated policy parameter. The first lane took the exact 2027 timing and England-outside-London scope from registered Table 6.1 line 4, and wrote in its row's note that IFS prints only "£2 cap on bus fares" on the slide. + +Conventions: +- We keep GBP amounts as IFS prints them, in cash or 2026-27 prices as it specifies, and multiply the £5bn aggregate by 1e9. +- We divide printed percentages by 100 for shares, while keeping CPI percentages and the printed 1% income decrease as percentages. +- We record a positive cost as a cost to the Exchequer, and positive household support as a gain. We don't re-sign IFS's printed 1% decrease in real household disposable income, and state its direction in sign_convention. +- For October 2026–June 2027 figures we use the annual time_basis, end-period marker 2027 and horizon strings, and no fiscal-year condition, because the nine months cross fiscal years. +- IFS's 1% RHDI decrease spans FY2024-25 to FY2026-27, and we use FY2026-27/end 2027. +- We describe no row as central: IFS prints point estimates and parameters throughout. +- For the gas reform IFS prints no implementation year, so we use 2026 as a publication marker only. +- We leave the baseline null and invented no unsupported counterfactual definitions. Not staged: -- Spending-decile bars in pp4–6, energy-price bars/percentiles pp7/16/17, carbon implicit-tax bars p9, and real-income decade/forecast lines pp11/14 have no data labels (apart from the separately staged£24/£36/£80 all-household labels and1% annotation). Visual inspection found no further printed values, embedded raster tables, or hidden numeric labels. -- Slide5 repeats p4's £140/£24/£36/£80 chart values; only its new80%/91% annotations are additional rows. Repeated£100 legend descriptions are not extra rows. -- Richest30% "lose a similar amount" is not an exact printed cash number; no negative£13 was inferred. --30% subgroup labels,5–15 bus-travel age eligibility, dates, axis ticks, years and0% chart origins are conditions/context, not separate scored estimates. -- Other temporary announcements on p15 (food tariffs, family-days-out VAT, free bus travel) have no printed cost/effect values, apart from the£2 bus fare parameter staged. -- Gas-standing-charge reform's average "no give-away" and "does not change overall costs" are qualitative; no numerical zero was invented. Its future electricity counterpart has no printed effect. -- No data download was identified within the PDF; the collection/event fetches were blocked as documented above. - -Potential contradictions/limits: the package's modelled full electricity-VAT period includes already-announced relief, despite being discussed beside Budget measures. This is why the package and full-period electricity rows have proposed scopes. The printed£140 headline is rounded; do not derive a different total from a plotted stack. Gas reform has unspecified implementation year and geography. The positive1% magnitude is explicitly a decrease. No numerical difference was detected between TW_0 and TW_1. - -Component spot check: deterministic random sample seed230,10 of21 rows, reopened directly from TW_1 PDF text; labels and direction were also checked against rendered pp4,5,7,8,14. One-based component rows1,6,7,8,10,11,14,15,18,19: p2 3.8% CPI inQ12027; p3 0% VAT October2026–June2027; p3 £5bn positive Exchequer cost across nine months; p4 £140 nine-month household support; p4 £36 VAT label; p4 £80 payment label (all in2026-27 prices); p7 around35% offset→0.35; p8 current£108 annual gas charge; p8 40% worse off→0.4 losing share; p8 poorest30% gain£13 per year, date unspecified/publication marker2026. All values found verbatim. The21 component rows pass row_problems with0 problems using UV_CACHE_DIR=/tmp/ab2026-ifs-uv-cache (default uv cache is outside the sandbox). +- IFS prints no data labels on the spending-decile bars in pp4–6, the energy-price bars/percentiles on pp7/16/17, the carbon implicit-tax bars on p9 or the real-income decade/forecast lines on pp11/14, apart from the separately staged £24/£36/£80 all-household labels and 1% annotation. The lane's visual inspection found no further printed values, embedded raster tables or hidden numeric labels. +- On slide 5 IFS repeats p4's £140/£24/£36/£80 chart values; we stage only its new 80%/91% annotations as further rows. We don't stage IFS's repeated £100 legend descriptions as extra rows. +- IFS says the richest 30% "lose a similar amount" without printing an exact cash number, and we inferred no negative £13. +- We treat the 30% subgroup labels, dates, axis ticks, years and 0% chart origins as conditions or context, not separate scored estimates. We stage no row with the 5–15 bus-travel age eligibility. +- IFS prints no cost or effect values for the other temporary announcements on p15 (food tariffs, family-days-out VAT, free bus travel), apart from the £2 bus fare parameter. +- IFS's "no give-away" and "does not change overall costs" for the gas-standing-charge reform are qualitative, and we invented no numerical zero. IFS prints no effect for its future electricity counterpart. +- The lane found no data download in the PDF, and IFS's server blocked the collection and event fetches, as documented above. + +Potential contradictions and limits: +- IFS's modelled full electricity-VAT period for the package includes relief announced earlier, though IFS discusses it beside Budget measures, so the first lane gave the package and full-period electricity rows proposed scopes. +- IFS rounds the printed £140 headline; don't derive a different total from a plotted stack. +- IFS leaves the gas reform's implementation year and geography unspecified. +- IFS gives the positive 1% magnitude explicitly as a decrease. +- The lane detected no numerical difference between TW_0 and TW_1. + +The lane drew a deterministic random sample, seed 230, of 10 of the 21 rows, and reopened each directly from the TW_1 PDF text. It also checked labels and direction against rendered pp4, 5, 7, 8 and 14. From the one-based component rows 1, 6, 7, 8, 10, 11, 14, 15, 18 and 19 it read: +- p2 3.8% CPI in Q1 2027; +- p3 0% VAT October 2026–June 2027; +- p3 £5bn positive Exchequer cost across nine months; +- p4 £140 nine-month household support; +- p4 £36 VAT label; +- p4 £80 payment label (all in 2026-27 prices); +- p7 around 35% offset → 0.35; +- p8 current £108 annual gas charge; +- p8 40% worse off → 0.4 losing share; +- p8 poorest 30% gain £13 per year, date unspecified/publication marker 2026. + +The lane found all values verbatim. The 21 component rows pass row_problems with 0 problems using UV_CACHE_DIR=/tmp/ab2026-ifs-uv-cache (the default uv cache lies outside the sandbox). ## Bee Boileau public services deck Primary: *Spending on public services: the Achilles’ heel?*, 29 October 2026, https://ifs.org.uk/system/files/2026-10/Spending-on-public-services-the-Achilles-heel-BB.pdf. -Fresh direct fetch 2026-10-29 15:43:43Z with the required Safari User-Agent; -157,838 bytes, sha256 `7403ecc0641b23f5ffc0cb28f67435dbd1b8df6b124874ffae7ae7f75e357aa6`. -The coordinator's assigned local copy and fresh copy are byte-identical. Read with -`pdftotext -layout`; visually inspected all 13 slides and the slide 7/9 data labels. -No raster-only numeric table was missed. - -Coverage: 28 rows, all `measure_key: null` (one has the proposed package below). -By slide: 3:1, 4:4, 5:1, 7:6, 8:1, 9:10, 10:3, 11:2. -The figures are 21 own arithmetic calculations/scenario assumptions and 7 restated -announced policy/target/reserve figures. No exact registered AB2026 measure is assigned. - -The first lane's one proposed key here, for the £1.2bn a year of DIP funding, went with that row, which the addendum dropped as an allocation (see "Keys the addendum changed" above). The rule 8 drops also removed slide 4's and 8's targets, slide 5's financing and the two scenario inputs on slides 7 and 9, so the deck now has 21 rows. - -Metric proposals reuse `spending_pressure`, `forecast_spending_change` and -`real_spending_growth` from AB2025. The historical one-percentage-point revision uses `real_spending_growth_revision`. New `reserve_allocated_share` means the -percentage of the specified day-to-day/capital reserve already allocated over -its printed multi-year window. No new unit or baseline proposal. - -`arithmetic` is the IFS's own scenario/growth calculation, or an announced target -restated without a named model; the £1.2bn funding amount is marked restated with -`hmt_costing_obr_certified`. Restated rows name their government/DIP origin in -`note`; two announced reserve allocation shares are `administrative_fact`, with -medium confidence, since the slide names no source for them. -Real growth remains percent, growth revisions are percentage points, cash -amounts are GBP, and signs are as printed. Positive pressure means additional -spending required; negative growth is a real cut. There are no central-value -rows: IFS's scenarios are point estimates, never substituted for OBR forecasts. - -The 2027–28 to 2030–31 chart rates are average annual real growth over the whole -interval: `growth_period` preserves that interval, and the row's fiscal period -is its terminal year, 2031. The reserve shares likewise retain their complete -allocation windows, ending 2028–29 and 2029–30 respectively. The published £1.2bn -and ~£1bn annual flows have no fiscal year: they carry a 2026 publication-date -snapshot, `time_basis: point_in_time`, `reference_date`, and medium confidence; -the rows do not claim these are 2026 annual outlays. The historical average -1-percentage-point growth revision since 2010 uses the same dated-snapshot -convention. Slide 4 states a calendar-2030 defence target, so that row has no fy; -slide 9 explicitly says 2030–31 and is recorded as fiscal. The difference in date -labels is preserved, not resolved by the harvest. - -Not staged: slide 3's unlabelled growth bars and axis ticks; all dates, document -address numbers and slide enumeration; qualitative/unquantified claims; the -second occurrence of the identical 3%-of-GDP scenario annotation within slide 9. -Slide 7 and slide 9 repeated printed labels are retained at their separate -locators, and slide 9's two health bars are separate scenario observations. -The £1bn financing flow is explicitly investment/financing, not military -purchases, matching the slide's caution. The deck identifies no downloadable -data file or distributional chart. No public-services primary was unreached. - -Validation: `row_problems` from the AB2026 checker, with registered keys and enums, -reported **28 rows, 0 row problems**. Full-family validation is recorded below. A full visual pass found no contradictory value. -The calendar/fiscal defence target wording and unspecified flow years above are -material limitations. - -Deterministic random spot-check (seed 230, 1-based indices within the Boileau component): +The lane fetched it fresh and directly at 2026-10-29 15:43:43Z with the required +Safari User-Agent: 157,838 bytes, sha256 +`7403ecc0641b23f5ffc0cb28f67435dbd1b8df6b124874ffae7ae7f75e357aa6`. The +fresh copy matches the coordinator's assigned local copy byte for byte. The +lane read it with `pdftotext -layout` and visually inspected all 13 slides and +the slide 7/9 data labels. It missed no raster-only numeric table. + +The first lane staged 28 rows (21 after the 3 November addendum), all +`measure_key: null` (one with the proposed package below). By slide: 3:1, 4:4, 5:1, 7:6, 8:1, 9:10, 10:3, 11:2. +IFS gives 21 own arithmetic calculations or scenario assumptions and 7 restated +announced policy, target or reserve figures. The first lane assigned no exact +registered AB2026 measure. + +The first lane's one proposed key here, for the £1.2bn a year of DIP funding, went with that row, which we dropped under the addendum as an allocation (see "Keys the addendum changed" above). With our rule 8 drops we also removed slide 4's and 8's targets, slide 5's financing and the two scenario inputs on slides 7 and 9, so we now stage 21 rows from the deck. + +For metrics the first lane reused `spending_pressure`, `forecast_spending_change` +and `real_spending_growth` from AB2025, and used `real_spending_growth_revision` +for the historical one-percentage-point revision. It proposed the new +`reserve_allocated_share` for the percentage of the specified day-to-day or +capital reserve already allocated over its printed multi-year window, and +proposed no new unit or baseline. Under the 3 November addendum we kept 21 of +the deck's rows (rows 54–74). We stage them under the proposed metrics +`real_spending_growth` (11 rows), `spending_pressure` (7), +`reserve_allocated_share` (2) and `real_spending_growth_revision` (1), and none +under `forecast_spending_change`. + +The first lane used `arithmetic` for IFS's own scenario and growth calculations, +or for an announced target IFS restates without naming a model. It marked the +£1.2bn funding amount restated, with `hmt_costing_obr_certified`, and in `note` +named each restated row's government/DIP origin. It marked the two announced +reserve allocation shares `administrative_fact`, with medium confidence, since +IFS names no source for them on the slide. It kept real growth in percent, +growth revisions in percentage points and cash amounts in GBP, with signs as +printed, and recorded additional spending required as positive pressure and a +real cut as negative growth. It staged no central-value rows: IFS gives its scenarios +as point estimates, and the lane never substituted them for OBR forecasts. + +IFS gives the 2027–28 to 2030–31 chart rates as average annual real growth over +the whole interval. The first lane kept that interval in `growth_period` and +set the row's fiscal period to its terminal year, 2031. It likewise kept the +reserve shares' complete allocation windows, ending 2028–29 and 2029–30 +respectively. IFS gives no fiscal year for the published £1.2bn and ~£1bn annual +flows, so the lane gave them a 2026 publication-date snapshot, +`time_basis: point_in_time`, `reference_date` and medium confidence, without +claiming them as 2026 annual outlays. It treated IFS's historical average +1-percentage-point growth revision since 2010 the same way. On slide 4 IFS +states a calendar-2030 defence target, so the first lane gave that row no fy; on +slide 9 IFS says 2030–31 explicitly, and the lane recorded that row as fiscal. +It kept the difference in date labels and didn't resolve it. + +The first lane didn't stage slide 3's unlabelled growth bars and axis ticks; +all dates, document address numbers and slide enumeration; qualitative or +unquantified claims; or the second occurrence of the identical 3%-of-GDP +scenario annotation within slide 9. It kept slide 7's and slide 9's repeated +printed labels at their separate locators, and staged slide 9's two health bars +as separate scenario observations. IFS describes the £1bn financing flow +explicitly as investment or financing, not military purchases, and the lane +followed the slide's caution. IFS identifies no downloadable data file or +distributional chart in the deck, and the lane reached every public-services +primary. + +`row_problems` from the AB2026 checker, with registered keys and +enums, reported **28 rows, 0 row problems**. We record full-family validation +below. In a full visual pass the lane found no contradictory value, and counted +the calendar/fiscal defence target wording and the unspecified flow years above +as material limitations. + +The lane's deterministic random spot-check (seed 230, 1-based indices within the Boileau component): | Row | Locator | Printed value | Normalized unit/value | Period check | |---|---|---|---|---| | 1 | slide 3 | £8 billion | GBP 8,000,000,000 | FY2030–31 | | 6 | slide 5 | ~£1bn per year | GBP 1,000,000,000 | no year printed; dated snapshot, medium | -| 10 | slide 7 | 1.7% | percent 1.7 | average 2027–28 to2030–31 | -| 14 | slide 9 | 0.9% | percent 0.9 | average 2027–28 to2030–31 | +| 10 | slide 7 | 1.7% | percent 1.7 | average 2027–28 to 2030–31 | +| 14 | slide 9 | 0.9% | percent 0.9 | average 2027–28 to 2030–31 | | 16 | slide 9 | £9 billion | GBP 9,000,000,000 | interval ends FY2030–31 | | 18 | slide 9 | 2.7% | percent 2.7 | yellow health bar, average interval | | 22 | slide 9 | -0.9% | percent -0.9 | green other-departments bar, average interval | | 23 | slide 9 | -2.1% | percent -2.1 | yellow other-departments bar, average interval | | 24 | slide 10 | £28 billion | GBP 28,000,000,000 | FY2030–31 | -| 25 | slide 10 | 1 percentage point | percentage_points 1 | historic since2010, dated snapshot | +| 25 | slide 10 | 1 percentage point | percentage_points 1 | historic since 2010, dated snapshot | -For every sampled row the primary PDF was reopened with `pdftotext -f/-l` and the -number/string, sign, unit and context checked; slide 7/9 labels were also viewed -at full-page resolution. All printed value strings matched. No bar reading or -arithmetic derived by the harvest was used. +For every sampled row the lane reopened the primary PDF with `pdftotext -f/-l` +and checked the number or string, sign, unit and context; it also viewed slide +7/9 labels at full-page resolution. All printed value strings matched. The lane +used no bar reading and no arithmetic of its own. ## Stuart Adam: Tax measures -18 rows from one six-slide PDF. Direct fetch with the required browser User-Agent, -2026-10-29 15:43:43Z; 133,005 bytes; -sha256 `6792dc448f40d34f174b6a3c839863b5c969a7334263f5d6b539be62a0428bb8`. -The fetched bytes match the assignment's local copy. Text read with -`pdftotext -layout`; all six slides rendered and visually read as a safeguard -against raster text. Slides 2–6 have no embedded raster images and all printed -numbers are in the text extraction. No distributional chart or data download -is linked inside this deck. +The first lane staged 18 rows (12 after the 3 November addendum) from one six-slide PDF. It fetched the PDF +directly with the required browser User-Agent at 2026-10-29 15:43:43Z: 133,005 +bytes, sha256 `6792dc448f40d34f174b6a3c839863b5c969a7334263f5d6b539be62a0428bb8`. +The fetched bytes match the assignment's local copy. The lane read the text with +`pdftotext -layout`, and rendered and read all six slides as a safeguard against +raster text. IFS embeds no raster images on slides 2–6, and the lane's text extraction +captured every printed number. IFS links no distributional chart or data download +inside this deck. -Rows by registered key: `ab2026__package_total_tax_policy_decisions` 1; +The first lane keyed its 18 rows: `ab2026__package_total_tax_policy_decisions` 1; `ab2026__writing_down_allowance_12pc_and_fya_50pc` 1; `ab2026__distributions_framework_modernised` 1; `ab2026__badr_lifetime_limit_500000` 1; `ab2026__founders_relief_cgt` 6; `ab2026__business_rates_sbrr_rhl_thresholds_raised` 3; `ab2026__electricity_vat_zero_rate_to_july_2027` 1; null 4. -Since 3 November, the £7bn net tax rise, Table 6.1's total of tax policy decisions, takes `decisions_effect_on_borrowing` (README rule 10). +We now key the 12 rows we stage (rows 75–86) to `ab2026__founders_relief_cgt` 4; +`ab2026__hmrc_tax_gap_package` 2; +`ab2026__package_property_and_savings_income_rates` 1; +`ab2026__package_total_tax_policy_decisions` 1; +`ab2026__writing_down_allowance_12pc_and_fya_50pc` 1; +`ab2026__distributions_framework_modernised` 1; +`ab2026__badr_lifetime_limit_500000` 1; and +`ab2026__business_rates_sbrr_rhl_thresholds_raised` 1. +Since 3 November we put the £7bn net tax rise, Table 6.1's total of tax policy decisions, on `decisions_effect_on_borrowing` (README rule 10). + ### Proposed measure scopes -Superseded on 3 November (see "Keys the addendum changed" above): the self-assessment timing and the compliance-staff and debt-collection figures are line 45 components, matched to Table 6.3's items by value, and the savings and property income figure uses `ab2026__package_property_and_savings_income_rates`. Rule 8 dropped this deck's six parameter rows (the Founders' Relief rate, comparator and limit, the RHL rateable values and the 0% VAT rate), so it now has 12 rows. +We superseded these on 3 November (see "Keys the addendum changed" above). We stage the self-assessment timing and the compliance-staff and debt-collection figures as line 45 components, matched to Table 6.3's items by value, and the savings and property income figure on `ab2026__package_property_and_savings_income_rates`. Under rule 8 we dropped this deck's six parameter rows (the Founders' Relief rate, comparator and limit, the RHL rateable values and the 0% VAT rate), so we now stage 12 rows from this deck. ### Conventions and metric proposals -All monetary figures preserve printed signs. Revenue-raising amounts are -positive yields; the Founders' Relief £0.3bn is a positive cost, explicitly -labelled as such. Amounts expressed in £bn/£m are multiplied by 1e9/1e6. -Tax rates remain percent. All quoted excerpts have fewer than 15 words and -contain their exact `value_raw`. - -The HMT-certified Budget costings are `restated` with -`source_model: hmt_costing_obr_certified`; rates and eligibility parameters -are `restated` announced-government design with `administrative_data`. -The deck does not cite a source for its beneficiary count or gain calculations; -these are presented as IFS `own`, `arithmetic`. Its maximum business-rate -saving is also IFS arithmetic. `point` is used for printed single estimates; -the £12,000–£21,000 benefiting-property range uses range endpoints. No -uncertainty range is inferred. There are no rows labelled `central` in this deck. +The first lane kept printed signs on all monetary figures. It staged +revenue-raising amounts as positive yields, and the Founders' Relief £0.3bn as a +positive cost, labelled as such. It multiplied amounts in £bn/£m by 1e9/1e6, +kept tax rates in percent, and kept every quoted excerpt under 15 words, +containing its exact `value_raw`. Under the 3 November addendum we kept 12 of +these rows (rows 75–86) with these signs, scalings and quotes, among them row +82's Founders' Relief £0.3bn as a positive cost, and dropped every tax-rate row +(rule 8). + +The first lane marked the HMT-certified Budget costings `restated` with +`source_model: hmt_costing_obr_certified`, and rates and eligibility parameters +`restated` announced-government design with `administrative_data`. IFS cites no +source for its beneficiary count or gain calculations, so the lane presented +them, and IFS's maximum business-rate saving, as IFS `own`, `arithmetic`. It +used `point` for printed single estimates and range endpoints for the +£12,000–£21,000 benefiting-property range, inferred no uncertainty range, and +labelled no row in this deck `central`. - Reused `tax_threshold_level` (GBP): the £10m cap on qualifying Founders' - Relief gains and the lower/upper rateable-value bounds identifying RHL - properties benefiting from higher relief thresholds. -- Proposed `statutory_tax_rate` (percent): the statutory 10% Founders' Relief rate, - its current 24% standard-rate comparator, and the temporary 0% electricity - VAT rate. These are statutory parameters, not effective marginal rates. + Relief gains, and the lower and upper rateable-value bounds of the RHL + properties that benefit from higher relief thresholds. +- Proposed `statutory_tax_rate` (percent): the statutory 10% Founders' Relief + rate, its current 24% standard-rate comparator, and the temporary 0% + electricity VAT rate. IFS gives statutory parameters here, not effective + marginal rates. - Proposed `taxpayer_gain` (GBP): the printed maximum £1.4m gain per Founders' Relief beneficiary, with the maximum recorded in conditions. -- Proposed `business_rates_bill_change` (GBP): maximum annual bill saving +- Proposed `business_rates_bill_change` (GBP): the maximum annual bill saving to an RHL property from the higher relief thresholds, £7,146 in 2027-28; - positive means a gain to the business. - -Fiscal-year impacts use their end year. The BADR £0.3bn bullet omits the -year; 2031–32 is inherited from the preceding slide's final-year comparison -and its adjacent Founders' Relief bullet. That row has medium parse confidence -and explains the inheritance. The annual count of 600 beneficiaries and -average £500,000 gain have no year printed; they use an annual 2026 -publication-year marker, with explicit `period_assignment` and `horizon` -conditions and medium confidence. This marker does not claim the relief operated in 2026. -Unspecified-year parameters and the maximum £1.4m gain use point-in-time -2026 as announcements, with no fiscal-year condition. No baseline proposal. + positive for a gain to the business. + +The first lane gave fiscal-year impacts their end year. IFS omits the year in +the BADR £0.3bn bullet, so the lane took 2031–32 from the preceding slide's +final-year comparison and the adjacent Founders' Relief bullet, and gave that +row medium parse confidence with the inheritance explained. IFS prints no year +for the annual count of 600 beneficiaries or the average £500,000 gain. The lane +used an annual 2026 publication-year marker for them, with explicit +`period_assignment` and `horizon` conditions and medium confidence; the marker +doesn't claim the relief operated in 2026. It staged parameters with no year, +and the maximum £1.4m gain, at point-in-time 2026 as announcements, with no +fiscal-year condition, and proposed no baseline. ### Omissions and source issues -- No effective marginal rates, distributions or unlabelled charts occur in - this deck. No amounts are inferred from qualitative bullets. -- Dates and durations are descriptive scope rather than numerical impact +- IFS prints no effective marginal rates, distributions or unlabelled charts in + this deck, and we infer no amounts from qualitative bullets. +- We treat dates and durations as descriptive scope, not numerical impact rows: the bank surcharge's 15-month period; two years of business-rate discounts; the 2029 revaluation; April 2028 income-tax revenue devolution. - The devolution slide prints no quantified spending or revenue amount. -- The title slide prints “28 October 2026”, despite the assignment describing - the 29 October briefing and PDF metadata showing creation/modification on - 29 October. Publication date follows the printed cover (2026-10-28); the - contradictory date evidence is preserved here, not silently corrected. -- IFS's £1.1bn combined savings/property yield is retained verbatim. The - corresponding Table 6.1 line-61/62 final-year amounts are £455m and £590m, - totalling £1.045bn. Different rounding/aggregation could explain the - displayed £1.1bn; no correction or derivation is staged. -- The £1bn HMRC compliance/debt-collection figure is narrower than line 45's - full £3.160bn. It is not treated as a contradictory costing of the full package. + IFS prints no quantified spending or revenue amount on the devolution slide. +- IFS prints “28 October 2026” on the title slide, though the coordinator + assigned the 29 October briefing, and in the PDF metadata IFS records creation + and modification on 29 October. We follow the printed cover for the publication + date (2026-10-28). +- We keep IFS's £1.1bn combined savings/property yield verbatim. In Table 6.1 + HM Treasury gives lines 61 and 62 final-year amounts of £455m and £590m, totalling + £1.045bn. Different rounding or aggregation could explain IFS's displayed + £1.1bn; we stage no correction or derivation. +- IFS's £1bn HMRC compliance/debt-collection figure covers less than line 45's + full £3.160bn, and we don't treat it as a contradictory costing of the full + package. ### Verification -The 18 individual rows pass `row_problems` from the harvest contract with -zero problems. All 18 short quotes were matched against whitespace-normalised -`pdftotext -layout` output and visually checked on the rendered slides. -Random spot-check seed 230 selected the following ten local rows. Each was -reopened at its slide and checked for exact value, sign, unit and period: +The 18 individual rows pass `row_problems` from the harvest contract with zero +problems. The lane matched all 18 short quotes against whitespace-normalised +`pdftotext -layout` output and checked them visually on the rendered slides. +Its random spot-check, seed 230, selected the following ten local rows. It +reopened each at its slide and checked the exact value, sign, unit and period: | Local row | Slide | Printed value | Sign/basis | Period | |--:|--:|---|---|---| @@ -574,8 +670,8 @@ reopened at its slide and checked for exact value, sign, unit and period: ## First lane's family verification (29 October 2026, 106 rows) -The required command, with its uv cache confined to the ignored family primaries -directory, completed successfully: +The lane ran the required command, with its uv cache confined to the ignored +family primaries directory, and it completed successfully: ```sh UV_CACHE_DIR=.ab2026-primaries/uk_ifs_ab2026/uv-cache PYTHONPATH=. uv run -q python pipeline/ab2026_harvest_contract.py sources/harvest-uk-ab2026-2026-10-29/uk_ifs_ab2026 @@ -585,19 +681,19 @@ UV_CACHE_DIR=.ab2026-primaries/uk_ifs_ab2026/uv-cache PYTHONPATH=. uv run -q pyt sources/harvest-uk-ab2026-2026-10-29/uk_ifs_ab2026: 106 rows, 0 problems ``` -Re-running `stage.py` produced the same gzip SHA256 before and after: +When the lane re-ran `stage.py`, it produced the same gzip SHA256 before and after: `4f0c9cd197d970486608ea093bf3822ae1d92fa56ebc4a2ac1fae3086430b1cb`. -68 text-row value_raw strings were also matched at their exact PDF slide -locators. Sixty-five quotes match the whitespace-normalised extraction directly; -three Boileau chart annotations interleave with other columns in pdftotext and -were confirmed against rendered slides 7 and 9. All 38 embedded fiscal-table -cells were independently checked visually. All quotes have fewer than 15 words -and contain value_raw. - -The final family random sample uses seed 230, ten of 106 rows, one-based gzip -indices. Each primary was reopened at the locator and its number, sign, unit and -period checked. All ten passed; unspecified dates are disclosed markers, not -confirmed implementation dates. +The lane also matched 68 text-row value_raw strings at their exact PDF slide +locators. Sixty-five quotes match the whitespace-normalised extraction directly. +In pdftotext three Boileau chart annotations interleave with other columns, so +the lane confirmed them against rendered slides 7 and 9. It checked all 38 +embedded fiscal-table cells independently by eye. Every quote runs under 15 +words and contains value_raw. + +For the final family random sample the lane used seed 230, ten of 106 rows, +one-based gzip indices. It reopened each primary at the locator and checked the +number, sign, unit and period. All ten passed; we disclose unspecified dates as +markers, not confirmed implementation dates. | Row | Primary / locator | Value and sign | Normalized unit | Period checked | |--:|---|---|---|---| @@ -612,7 +708,7 @@ confirmed implementation dates. | 93 | Adam, slide 2, HMRC staffing/debt collection | £1bn, positive yield | GBP 1,000,000,000 | FY2031-32 / 2032 | | 106 | Adam, slide 5, electricity VAT | 0%, statutory rate | percent 0 | announcement at publication in 2026; no fy | -Registered-key scope review covered every non-null row: +The lane reviewed the registered-key scope of every non-null row: | Registered key suffix (all prefixed ab2026__) | Table 6.1 line | Printed scope checked | |---|--:|---| @@ -627,11 +723,10 @@ Registered-key scope review covered every non-null row: | electricity_vat_zero_rate_to_july_2027 | 2 | Adam statutory zero-rate parameter, no full-period gain/cost attached | | package_total_tax_policy_decisions | total | net Budget tax rise in final forecast year | -Wernham's full-period VAT *impact* uses a proposal because it includes prior -relief; Adam's zero-rate *parameter* is directly stated for the measure and is -not a costing of the full period. All broader or unidentified packages use null -measure_key with proposals. Source issues are preserved: Adam's cover date, +The first lane gave Wernham's full-period VAT *impact* a proposal because it +includes earlier relief; since 3 November we key it to line 2. Adam states the +zero-rate *parameter* directly for the measure and doesn't cost the full period. +The first lane gave all broader or unidentified packages a null measure_key with +proposals. It left these as IFS printed them: Adam's cover date, the savings/property rounding, Boileau's calendar/fiscal target wording and undated -flows, Wernham's undated gas reform, and Stockton's uncited attribution. None -was silently corrected. - +flows, Wernham's undated gas reform, and Stockton's uncited attribution. diff --git a/sources/harvest-uk-ab2026-2026-10-29/uk_ippr/NOTES.md b/sources/harvest-uk-ab2026-2026-10-29/uk_ippr/NOTES.md index 7930900a0..53e98d67f 100644 --- a/sources/harvest-uk-ab2026-2026-10-29/uk_ippr/NOTES.md +++ b/sources/harvest-uk-ab2026-2026-10-29/uk_ippr/NOTES.md @@ -1,25 +1,25 @@ # uk_ippr: IPPR, Autumn Budget 2026 -This family stages IPPR's three responses to the Budget on 28 October 2026, found on the ippr.org media-office listing: +In this family we stage IPPR's three responses to the Budget on 28 October 2026, which we found on the ippr.org media-office listing: - "‘Unfinished business’ on economic reform after sound Budget, says IPPR"; - "IPPR North reacts to Andy Burnham’s first Budget"; - "IPPR Scotland reacts to Andy Burnham's first Budget". -The listing shows nothing on 29 October. The source slug is `ippr`, as in the AB2025 `uk_ippr` family. +IPPR lists nothing on 29 October. We use the source slug `ippr`, as in the AB2025 `uk_ippr` family. -**Working files.** `stage.py`, the file meant wherever these notes say `stage.py`, lives on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/thinktanks-a/uk_ippr/stage.py`; this branch carries only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz`. The lane's local primaries folder (`.ab2026-primaries/uk_ippr/`) was never committed: the repository carries no primary bytes, so fetch each primary from its `manifest.jsonl` URL and check its sha256. +We keep `stage.py` on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/thinktanks-a/uk_ippr/stage.py`. We commit only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz` to this branch, and no primary bytes to the repository. The lane never committed its local primaries folder (`.ab2026-primaries/uk_ippr/`), so fetch each primary from its `manifest.jsonl` URL and check its sha256. -`stage.py` holds the rows as data and rewrites `manifest.jsonl` and `claims_staged.jsonl.gz` byte for byte. `stage.py --verify` checks each primary's sha256 and finds every quote in its page's text. Result: 0 problems. +We wrote the rows into `stage.py` as data, and it rewrites `manifest.jsonl` and `claims_staged.jsonl.gz` byte for byte. `stage.py --verify` checks each primary's sha256 and finds every quote in its page's text; it reports 0 problems. -- **Rows:** 3 (restated 2, own 1). -- **Primaries:** 3. +- Rows: 3 (restated 2, own 1). +- Primaries: 3. ## Access recipe -- **The listing:** the media-office listing (https://www.ippr.org/media-office, pages 1 and 2) was fetched on 2026-10-29 to find the responses. `/news-and-media` returns 404, and `/articles` lists no Budget piece. -- **The releases:** all three were fetched directly (`curl -sSL`, browser User-Agent, HTTP 200), "Unfinished business" at 16:17Z and the North and Scotland releases at 16:35Z. -- **`dateModified` is a render time, not an edit:** each page's JSON-LD showed a `dateModified` on 29 October (16:04Z, 16:21Z, 16:34Z), after its 28 October `datePublished`. Review B's fetches on 4 November show 14:39–14:46Z that day with byte-identical length, and two fetches of the same page 30 minutes apart differ only in that field and Cloudflare's email tokens. So the field records when the page was served, and no edit after publication is shown. The copies hashed are the fetches above. Nothing came from a local copy. +- We fetched the media-office listing (https://www.ippr.org/media-office, pages 1 and 2) on 2026-10-29 to find the responses. IPPR's `/news-and-media` returns 404, and IPPR lists no Budget piece on its `/articles`. +- We fetched all three releases directly (`curl -sSL`, browser User-Agent, HTTP 200): "Unfinished business" at 16:17Z and the North and Scotland releases at 16:35Z. +- We read each page's JSON-LD `dateModified` as a render time, not an edit. On 29 October IPPR's server gave each page a `dateModified` that day (16:04Z, 16:21Z, 16:34Z), after its 28 October `datePublished`. Review B's fetches on 4 November show 14:39–14:46Z that day with byte-identical length. Two fetches of the same page 30 minutes apart differ only in that field and Cloudflare's email tokens. So IPPR's server stamps the field when it serves the page, and we found no sign of an edit after publication. We hashed the copies from the fetches above and took nothing from a local copy. ## Rows by primary and measure key @@ -31,32 +31,33 @@ The listing shows nothing on 29 October. The source slug is `ippr`, as in the AB | measure_key | rows | |---|--:| -| `null` (1 with a proposed key, 2 context) | 3 | +| `ab2026_option__income_tax_5pc_share_devolved_to_mayors` | 1 | +| `null` (context) | 2 | ## The three rows 1. **"It will lower inflation by at least 0.2 percentage points"** (Carsten Jung, on "the cost of living support announced today"). - - Staged as `cpi_inflation_effect`, 0.2, `range_low`, with `sign_convention: "as worded: positive = reduction in CPI inflation"`: IPPR prints the magnitude unsigned, README rule 10 keeps the producer's sign, and "at least" makes it a lower bound (changed on 4 November, review B finding 1). Context, with no proposed key: IPPR's words do not define the scope (changed on 3 November, thinktanks-a review finding 1). - - IPPR names no source and no period. The OBR's EFO paragraph 3.10 says the fuel duty changes and the temporary VAT cut on domestic electricity (Table 6.1 lines 2 and 3) "reduce CPI inflation by 0.22 percentage points in the second quarter of 2027". So the row is `restated` with `parse_confidence: medium`, at `period` 2027 (`point_in_time`). + - We stage it as `cpi_inflation_effect`, 0.2, `range_low`, with `sign_convention: "as worded: positive = reduction in CPI inflation"`. IPPR prints the magnitude unsigned, under README rule 10 we keep the producer's sign, and "at least" makes it a lower bound; we made this change on 4 November, after review B finding 1. We stage it as context, with no proposed key, because IPPR's words do not define the scope; we made that change on 3 November, after thinktanks-a review finding 1. + - IPPR names no source and no period. In the EFO's paragraph 3.10, the OBR says the fuel duty changes and the temporary VAT cut on domestic electricity (Table 6.1 lines 2 and 3) "reduce CPI inflation by 0.22 percentage points in the second quarter of 2027". We therefore mark the row `restated`, with `parse_confidence: medium`, at `period` 2027 (`point_in_time`). 2. **"the £18 billion annual taxpayer losses at the Bank of England"** (Carsten Jung). - - Staged as `apf_indemnity_losses`, the AB2025 uk_ippr proposal, which staged IPPR's "£22 billion" the same way. `restated`, source not stated. - - It has no year, so it sits at the scoring year 2026-27 with a `year_note`. This is context with no measure key. + - We stage it as `apf_indemnity_losses`, the AB2025 uk_ippr proposal, under which we staged IPPR's "£22 billion" the same way. We mark it `restated`, source not stated. + - Jung gives no year, so we put it at the scoring year 2026-27 with a `year_note`. We stage it as context with no measure key. 3. **"Five per cent of income tax devolved to mayors would see an additional £4billion"** (IPPR North). - - IPPR North's own estimate of an option beyond the Budget's commitment on income tax devolution. Proposed key `ab2026_option__income_tax_5pc_share_devolved_to_mayors`, proposed metric `devolved_revenue` (gbp). - - It has no year, so it sits at the scoring year with a `year_note`. The release links the paper "Devolution that delivers: Fiscal power for regional growth". + - IPPR North estimates, on its own account, an option beyond the Budget's commitment on income tax devolution. We key it to `ab2026_option__income_tax_5pc_share_devolved_to_mayors` and propose the metric `devolved_revenue` (gbp). On 29 October we gave the row `measure_key: null` and proposed this key, and in tranche 2 we registered it. + - IPPR North gives no year, so we put it at the scoring year with a `year_note`. IPPR North links the paper "Devolution that delivers: Fiscal power for regional growth" from the release. -## Proposed keys and their scope +## Option keys and their scope -- **`ab2026_option__income_tax_5pc_share_devolved_to_mayors`:** "Five per cent of income tax devolved to mayors". No Table 6.1 line. The Budget made commitments "on devolving business rates and a share of income tax to England's mayors" (line 21 covers the business rates share) without setting a 5% share. +- **`ab2026_option__income_tax_5pc_share_devolved_to_mayors`:** "Five per cent of income tax devolved to mayors". HM Treasury prints no Table 6.1 line for it. The Budget made commitments "on devolving business rates and a share of income tax to England's mayors" (line 21 covers the business rates share) without setting a 5% share. ## Not staged, and why -- **Design, not estimates** (README rule 8: spending allocations, financing remits and payment amounts are a measure's design): - - IPPR North's "£285million of good growth funds" is HM Treasury's allocation ("£285 million in 2027-28, comprising £165 million to support places with a mayor … to establish Good Growth Funds and £120 million of incubator capital funding for areas transitioning towards a devolution deal"), restated. It is a spending allocation, and part of Table 6.1 line 20, whose costing (−£340m in 2027-28) also includes the subsidy on the £1.5bn finance. The reworked uk_hmt_redbook family does not stage HM Treasury's £285m either (rule 8). - - IPPR North's "£1.5billion borrowing facilities for building infrastructure" is the size of HM Treasury's lending facility for Mayoral Strategic Authorities, a financing remit; the reworked uk_hmt_redbook family stages no `policy_parameter` row and does not stage it either. - - IPPR Scotland's "Bill support of £100 for households in lower council tax bands" is the Household Energy Payment's amount. -- **IPPR Scotland** prints no other number, so it gives 0 rows; it is manifested for the tally. -- **Pre-Budget IPPR releases about this Budget,** outside the assigned "responses of 28–29 October". Each was fetched and hashed on 2026-10-29, and each is a candidate for a follow-up if the family is to include pre-Budget options, as AB2025's did: +- Under README rule 8 we treat spending allocations, financing remits and payment amounts as a measure's design, and left these out: + - IPPR North's "£285million of good growth funds" restates HM Treasury's allocation ("£285 million in 2027-28, comprising £165 million to support places with a mayor … to establish Good Growth Funds and £120 million of incubator capital funding for areas transitioning towards a devolution deal"). HM Treasury makes it a spending allocation and costs it within Table 6.1 line 20, whose costing (−£340m in 2027-28) also includes the subsidy on the £1.5bn finance. In the reworked uk_hmt_redbook family we don't stage HM Treasury's £285m either (rule 8). + - IPPR North restates the size of HM Treasury's lending facility for Mayoral Strategic Authorities, a financing remit, as "£1.5billion borrowing facilities for building infrastructure". In the reworked uk_hmt_redbook family we stage no `policy_parameter` row and leave it out there too. + - In "Bill support of £100 for households in lower council tax bands", IPPR Scotland restates the Household Energy Payment's amount. +- IPPR Scotland prints no other number, so we stage 0 rows from it; we manifest it for the tally. +- We left out IPPR's pre-Budget releases about this Budget, which fall outside the assigned "responses of 28–29 October". We fetched and hashed each on 2026-10-29, and each could seed a follow-up if the family should include pre-Budget options, as AB2025's did: - 27 October, "Chancellor’s £20bn to invest would cost less than £1 billion in headroom: IPPR analysis" (sha256 `e26f9b14989f14af…`): £20bn of NWF and BBB investment costing under £1bn in debt interest and running costs in 2029-30. - 15 October, "Get started with tax reform to pay for priorities…" (sha256 `97750268ba93a3b6…`): - a £16bn package: CGT reform about £10bn, deferring Bank of England losses £5bn, a bank windfall surcharge £1bn; @@ -71,18 +72,18 @@ The listing shows nothing on 29 October. The source slug is `ippr`, as in the AB - multipliers cut from 38.2p and 43p to 28.2p and 41p; - property taxes at 10.5% of revenue against an OECD average of 5.1%; - business rates of £27.6bn. - The papers behind them were not fetched: "Cost of living, economic growth and stability: How to make this autumn budget a success", "Bricks vs clicks", "Devolution that delivers" and "Beyond headroom". + We didn't fetch the papers behind them: "Cost of living, economic growth and stability: How to make this autumn budget a success", "Bricks vs clicks", "Devolution that delivers" and "Beyond headroom". ## Review fixes, 3 November 2026 -The thinktanks-a review (`harvest-staging/ab2026/reviews/verify-thinktanks-a.md` on the staging branch) approved this family with one fix, applied in an explicit block at the end of `stage.py`'s rows: +In its review (`harvest-staging/ab2026/reviews/verify-thinktanks-a.md` on the staging branch), the thinktanks-a reviewer approved this family with one fix, which we applied in an explicit block at the end of `stage.py`'s rows: -- **Finding 1:** row 0 (−0.2pp CPI) is context. Jung says only "The cost of living support announced today", and Quilter-Pinner "action to support families with the cost of living". The lines 2 and 3 scope came from matching the OBR's paragraph 3.10, which is our inference, and README rule 3 says the scope must be the producer's. So the proposed key `ab2026__package_cost_of_living_support` is withdrawn, the reform hint and note stay, and `value_kind` is `range_high`. -- **Verification:** no primary was re-opened (the session's network policy refused ippr.org); the contract checker reports 3 rows, 0 problems, and a re-run of `stage.py` reproduces both files byte for byte. +- Finding 1: we made row 0 (−0.2pp CPI) context. Jung says only "The cost of living support announced today", and Quilter-Pinner "action to support families with the cost of living". We had taken the lines 2 and 3 scope by matching the OBR's paragraph 3.10, our own inference, and under README rule 3 we must use the producer's scope. So we withdrew the proposed key `ab2026__package_cost_of_living_support`, kept the reform hint and note, and set `value_kind` to `range_high`. +- We re-opened no primary, because the session's network policy refused ippr.org. The contract checker reports 3 rows, 0 problems, and when we re-run `stage.py` it reproduces both files byte for byte. ## Review fixes, 4 November 2026 -Review B (PR #231, issue comment 6635498758) approved this family with fixes; review A's K-7(f) asked for the same NOTES correction as its second item. This session re-opened no primary. +Review B (PR #231, issue comment 6635498758) approved this family with fixes, and review A's K-7(f) asked for the same NOTES correction as its second item. We re-opened no primary in this session. -- **Finding 1 (coordinator's call: take the fix):** row 0 stores 0.2, `range_low`, `sign_convention: "as worded: positive = reduction in CPI inflation"`, as the OBR and red book CPI rows keep the magnitude (`uk_obr_tables` row 2653). -- **Finding 2 and K-7(f):** "Not staged" no longer says the red book stages HM Treasury's £285m or the £1.5bn as `policy_parameter`; the reworked red book stages neither. The access recipe reads JSON-LD `dateModified` as a render time, not an edit. +- Finding 1, where the coordinator chose to take the fix: on row 0 we now store 0.2, `range_low`, `sign_convention: "as worded: positive = reduction in CPI inflation"`. We keep the magnitude, as we do on the OBR and red book CPI rows (`uk_obr_tables` row 2653). +- Finding 2 and K-7(f): under "Not staged" we no longer say that the red book family stages HM Treasury's £285m or the £1.5bn as `policy_parameter`; in the reworked red book family we stage neither. In the access recipe we now read JSON-LD `dateModified` as a render time, not an edit. diff --git a/sources/harvest-uk-ab2026-2026-10-29/uk_obr_tables/NOTES.md b/sources/harvest-uk-ab2026-2026-10-29/uk_obr_tables/NOTES.md index 13381161f..8e7626c4d 100644 --- a/sources/harvest-uk-ab2026-2026-10-29/uk_obr_tables/NOTES.md +++ b/sources/harvest-uk-ab2026-2026-10-29/uk_obr_tables/NOTES.md @@ -1,16 +1,18 @@ # uk_obr_tables — October 2026 Economic and fiscal outlook -**Working files.** `stage.py`, the file meant wherever these notes say `stage.py`, lives on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/obr-tables/uk_obr_tables/stage.py`; this branch carries only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz`. The repository carries no primary bytes: fetch each primary from its `manifest.jsonl` URL (or, while obr.uk refuses non-browser clients, the coordinator's local copies) and check its sha256. +We keep `stage.py` on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/obr-tables/uk_obr_tables/stage.py`. We commit only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz` to this branch, and no primary bytes to the repository. Fetch each primary from its `manifest.jsonl` URL (or, while obr.uk refuses non-browser clients, use the coordinator's local copies) and check its sha256. -Staged 2026-10-29; reworked 3 November 2026 under the OBR review, and on 4 November under reviews A and B (see "Review fixes" at the end). Source `obr_efo`. **2,800 rows (2,891 before the 3 November review, 2,797 at 4f424b9), 4 primaries (rows cite 3 of them), 98 individual registered measures plus the registered total-policy package.** All outputs are confined to this family. `stage.py` embeds all reviewed rows as metadata templates and cell data; rerunning needs only the Python standard library and reproduces the gzip without primary files or network access. +We staged this family on 2026-10-29, reworked it on 3 November 2026 under the OBR review, and again on 4 November under reviews A and B (see "Review fixes" at the end). We use source `obr_efo`. We stage **2,800 rows (2,891 before the 3 November review, 2,797 at 4f424b9), 4 primaries (we cite 3 of them in rows), 98 individual registered measures plus the registered total-policy package.** We write all outputs inside this family. `stage.py` embeds all reviewed rows as metadata templates and cell data. A rerun needs only the Python standard library, and reproduces the gzip without primary files or network access. ## Access recipe and primaries -Each resolved URL in the supplied OBR `PROVENANCE.md` was attempted once using `curl -sSL` and the exact Safari User-Agent requested in the assignment. The EFO landing page returned **403** at 2026-10-29T15:51:17Z; the chapter-2 workbook and detailed-table ZIP resolved URLs returned **403** at 15:51:18Z. No bypass was attempted. We used the supplied local policy and aggregate workbook members and PDF; every row and manifest entry records `access: "local copy"`. +The lane tried each resolved URL in the supplied OBR `PROVENANCE.md` once, with `curl -sSL` and the exact Safari User-Agent the assignment requested. The OBR's server returned **403** for the EFO landing page at 2026-10-29T15:51:17Z, and for the resolved URLs of the chapter-2 workbook and detailed-table ZIP at 15:51:18Z. The lane attempted no bypass. We used the supplied local policy and aggregate workbook members and PDF, and record `access: "local copy"` on every row and manifest entry. -The coordinator supplied the local copies. The lane kept exact byte copies, failed-response headers and bodies and fetch timestamps in its local primaries folder, never committed. Workbooks were read with openpyxl `data_only=True`; the PDF with `pdftotext -layout`. `shasum -a 256` verified all three primaries. Both workbook member hashes match the browser-download provenance from 28 October, including the ZIP's recorded Last-Modified of Wed, 28 Oct 2026 14:05:59 GMT. +The coordinator supplied the local copies. The lane kept exact byte copies, failed-response headers and bodies, and fetch timestamps in its local primaries folder, which it never committed. It read the workbooks with openpyxl `data_only=True` and the PDF with `pdftotext -layout`, and verified all three primaries with `shasum -a 256`. Both workbook member hashes match the browser-download provenance from 28 October, including the ZIP's recorded Last-Modified of Wed, 28 Oct 2026 14:05:59 GMT. -The resolved ZIP URL is the publication URL for both members, with the member name in each row's note and the manifest. The ZIP container itself was not supplied locally and is not represented as if it had been read. The PDF attachment URL is absent from PROVENANCE; the blocked landing page prevented discovering it. Its manifest and rows therefore use the EFO landing URL and identify the supplied local PDF filename. The cover/imprint say October 2026; publication date 2026-10-28 follows the EFO-page date recorded in PROVENANCE. **The PDF changed after the lane read it (README rule 7).** GOV.UK replaced the attachment on 30 October 2026 (change history 2026-10-30T14:21:10Z) with a corrected version (sha256 `555e0c26…`, 1,771,863 bytes), and the OBR's download link (`obr.uk/download/economic-and-fiscal-outlook-october-2026/`) redirects to it. It adds a correction slip at physical p.4, and paragraph 3.19's first bullet now dates the abolition of carbon price support April 2028, where the 28 October text said April 2027. Review B found every other staged PDF quote unchanged in it. Its physical page is printed page + 4; the 28 October file's is printed + 3. The manifest lists it as a fourth primary; every PDF row cites the 28 October version (`9e0e9c34…`), which the lane read. +We give the resolved ZIP URL as the publication URL for both members, with the member name in each row's note and the manifest. Nobody supplied the ZIP container itself locally, and we don't present it as read. We found no PDF attachment URL in PROVENANCE, and couldn't discover it from the blocked landing page. So in the PDF's manifest entry and rows we use the EFO landing URL and name the supplied local PDF file. The OBR prints October 2026 on the cover and imprint; we take the publication date 2026-10-28 from the EFO-page date we recorded in PROVENANCE. + +**The OBR corrected the PDF after the lane read it (README rule 7).** On 30 October 2026 GOV.UK replaced the attachment (change history 2026-10-30T14:21:10Z) with a corrected version (sha256 `555e0c26…`, 1,771,863 bytes), and the OBR's download link (`obr.uk/download/economic-and-fiscal-outlook-october-2026/`) redirects to it. The OBR added a correction slip at physical p.4. In paragraph 3.19's first bullet it now dates the abolition of carbon price support April 2028, where in the 28 October text it said April 2027. Review B found every other staged PDF quote unchanged in it. Its physical page equals the printed page + 4, against printed + 3 in the 28 October file. We list it in the manifest as a fourth primary, and in every PDF row we cite the 28 October version (`9e0e9c34…`), which the lane read. | Primary file | Rows | Bytes | SHA-256 | |---|---:|---:|---| @@ -21,23 +23,31 @@ The resolved ZIP URL is the publication URL for both members, with the member na ## Coverage -- Policy workbook Table 1.1: **2,520** cells, every numeric row series (420) across all six fiscal years, 2026-27 through 2031-32. This includes 2,112 numbered-measure component/total cells, 48 fiscally neutral cells, 30 non-scorecard cells and 330 summary cells. -- EFO chapter 3: **225** rows (316 before the 3 November review): Table 3.1 114 (its "Spending measures" and "Tax measures" rows repeat its "Direct effect" rows and were dropped); Table 3.3 35 costings plus 7 categorical uncertainty ratings; policy/behavioural prose and Box 3.1 69 (paragraph 3.20's financing requirement restored and Box 3.1's two measure counts added on 4 November). Table 3.2's 47 funding cells are design (rule 8) and are not staged. -- Headline fiscal context: **55** rows: 13 workbook cells (PSNB cash six years; ex-BoE PSND/GDP seven years) and 42 PDF cells (Table 7.1 headroom 13; Annex A Table A.9 PSNB, PSNFL and ex-BoE PSND 29). The current-budget and PSNFL rules include current pre-measures, post-measures and the printed memo excluding fuel-duty rises; welfare-cap margin is included. Forecast duplicates between the PDF and aggregate workbook are represented once. +- Policy workbook Table 1.1: we stage **2,520** cells, every numeric row series (420) across all six fiscal years, 2026-27 through 2031-32. They comprise 2,112 numbered-measure component/total cells, 48 fiscally neutral cells, 30 non-scorecard cells and 330 summary cells. +- EFO chapter 3: we stage **225** rows (316 before the 3 November review): + - Table 3.1, 114 rows. The OBR repeats the table's "Direct effect" rows as "Spending measures" and "Tax measures" rows, so we dropped the repeats. + - Table 3.3, 35 costings plus 7 categorical uncertainty ratings. + - Policy and behavioural prose and Box 3.1, 69 rows. On 4 November we restored paragraph 3.20's financing requirement and added Box 3.1's two measure counts. + - We treat Table 3.2's 47 funding cells as design (rule 8) and don't stage them. +- Headline fiscal context: we stage **55** rows: 13 workbook cells (PSNB cash six years; ex-BoE PSND/GDP seven years) and 42 PDF cells (Table 7.1 headroom 13; Annex A Table A.9 PSNB, PSNFL and ex-BoE PSND 29). For the current-budget and PSNFL rules we include current pre-measures, post-measures and the printed memo excluding fuel-duty rises, and we include the welfare-cap margin. Where the OBR gives the same forecast in the PDF and the aggregate workbook, we stage it once. ## Conventions -- **Values and precision:** policy workbook values use their displayed integer £million formats (`0`, `#,##0`; eight exact General-format zeros), multiplied by 1e6. Unprinted cached decimal precision is not a claim. Decimal rounding is half away from zero; a stored negative fraction that displays zero is `0`. A LibreOffice PDF rendering independently confirmed these display conventions. Aggregate workbook values use displayed one-decimal precision. PDF £billion values multiply by 1e9 and £million by 1e6; no ratios, sums or inferred annual paths are staged. -- **Signs:** policy workbook prints “This table uses the convention that a negative sign implies a loss to the Exchequer (and is therefore an increase in PSNB).” PDF Tables 3.1/3.3 print “A positive sign implies an increase in borrowing.” Both are retained. The indirect-effect workbook row's conflict with that workbook legend is explicitly noted in those rows. Prose uses positive magnitudes for costs, yields or behavioural erosion exactly as worded; per-row conditions state the meaning. Headroom is positive margin, negative breach. -- **Period:** FY end year, with the printed `fy`; quarterly CPI peaks use calendar-year `point_in_time` and quarter detail without `fy`. An annual average or recurring prose amount appears once at the terminal forecast FY, with its interval in `aggregation`, rather than manufacturing six yearly values. Staffing figures are design (rule 8) and sit in `conditions.staffing`, not in rows. Box 3.1's undated share and counts sit at the publication-year anchor, 2026. Debt stocks preserve end-March timing and end-March-centred GDP denominators. -- **Central:** ordinary policy tables/prose are `point` estimates of the single final forecast; aggregate forecasts are `central`, meaning the OBR central forecast, and observed 2025-26 outturns are `point`. No statistical confidence interval is inferred. “High” ratings use categorical `value: 0` as the precedent's schema placeholder, not a numerical score. Health Foundation range endpoints span alternative reform options, not uncertainty for one policy. -- **Attribution:** 2,777 `own`, 23 `restated` (2,821 and 70 before Table 3.2's 47 rows went on 3 November). OBR-certified/HMRC/DWP-scrutinised costings follow the AB2025 `own` convention; OBR fiscal judgments use `obr_efo_forecast`. Twelve workbook fuel cells and two PDF fiscal fuel costings are restated `hmt_costing_uncertified`: EFO p.59 footnote 1 says the final Government costing arrived after certification. Six latest ONS outturn aggregate cells name ONS. Two social-care options name Health Foundation; one FTA effect names the Government/DBT impact assessment. OBR's CPI judgments remain its own estimates. ONS facts use `administrative_fact`; modeled estimates use `different_model`. -- **Scope:** all 98 numbered workbook titles match registry titles after whitespace/footnote normalization. Fiscal-head components retain their full measure's key, with component conditions; totals and components must not be summed as independent estimates. Only workbook row 552 and the PDF's explicit HMT-policy-table memo use the registered total-policy package. All other groupings use null plus a proposed key. Context levels use null without a proposed measure. Paragraph 3.28's salary-sacrifice revision is a context row with a null key (4 November, the coordinator's call on review A's H-6(b)); it is not recast as an AB2026 intervention. The AB2025 OBR family actually contains no headline fiscal aggregates; the live `Metric` enum supplies `fiscal_headroom`, and cross-family proposals supply the closest aggregate names. -- **Quotes/order:** full numeric/rating cells or full OGL prose sentences, preserving the printed number; PDF whitespace/line wraps normalized, footnote superscripts excluded from sentence text. Reading order is policy workbook row/column, PDF page/position, then aggregate workbook table/row/column. PDF printed page + 3 is the physical page of the 28 October file that rows cite (+ 4 in the 30 October correction). +- **Values and precision.** We take policy workbook values at their displayed integer £million formats (`0`, `#,##0`; eight exact General-format zeros) and multiply them by 1e6. We don't claim unprinted cached decimal precision. We round decimals half away from zero, and stage `0` for a stored negative fraction that displays zero. We confirmed these display conventions independently on a LibreOffice PDF rendering. We take aggregate workbook values at their displayed one-decimal precision. We multiply PDF £billion values by 1e9 and £million by 1e6, and stage no ratios, sums or inferred annual paths. +- **Signs.** The OBR prints in the policy workbook “This table uses the convention that a negative sign implies a loss to the Exchequer (and is therefore an increase in PSNB).” Under PDF Tables 3.1/3.3 it prints “A positive sign implies an increase in borrowing.” We keep both. On the workbook's indirect-effect rows, whose signs conflict with that workbook legend, we say so. In prose rows we keep positive magnitudes for costs, yields or behavioural erosion as the OBR words them, and state the meaning in per-row conditions. We record headroom as positive for a margin and negative for a breach. +- **Period.** We use the FY end year, with the printed `fy`. For quarterly CPI peaks we use calendar-year `point_in_time` and quarter detail without `fy`. We stage an annual average or recurring prose amount once, at the terminal forecast FY, with its interval in `aggregation`, rather than manufacture six yearly values. We treat staffing figures as design (rule 8) and keep them in `conditions.staffing`, not in rows. We put Box 3.1's undated share and counts at the publication-year anchor, 2026. We keep end-March timing for debt stocks and end-March-centred GDP denominators. +- **Central.** We stage ordinary policy tables and prose as `point` estimates of the single final forecast. We mark aggregate forecasts `central`, for the OBR central forecast, and observed 2025-26 outturns `point`. We infer no statistical confidence interval. For “High” ratings we use categorical `value: 0`, the precedent's schema placeholder, not a numerical score. The Health Foundation's range endpoints span alternative reform options, not uncertainty about one policy. +- **Attribution.** We mark 2,777 rows `own` and 23 `restated` (2,821 and 70 before we dropped Table 3.2's 47 rows on 3 November). + - We follow the AB2025 `own` convention for OBR-certified and HMRC- or DWP-scrutinised costings, and use `obr_efo_forecast` for the OBR's fiscal judgments. + - We mark twelve workbook fuel cells and two PDF fiscal fuel costings restated `hmt_costing_uncertified`: in EFO p.59 footnote 1 the OBR says the final Government costing arrived after certification. + - We name ONS on six latest ONS outturn aggregate cells, the Health Foundation on two social-care options, and the Government/DBT impact assessment on one FTA effect. + - We keep the OBR's CPI judgments as its own estimates. We use `administrative_fact` for ONS facts and `different_model` for modelled estimates. +- **Scope.** All 98 numbered workbook titles match registry titles once we normalise whitespace and footnotes. We key fiscal-head components to their full measure, with component conditions; never sum totals and components as independent estimates. We use the registered total-policy package only for workbook row 552 and the PDF's explicit HMT-policy-table memo. For all other groupings we use the package keys we proposed and registered in tranche 2, and for context levels null with no key. Since 4 November, by the coordinator's call on review A's H-6(b), we stage paragraph 3.28's salary-sacrifice revision as a context row with a null key, not as an AB2026 intervention. The AB2025 OBR family stages no headline fiscal aggregates. We take `fiscal_headroom` from the live `Metric` enum, and the closest aggregate names from cross-family proposals. +- **Quotes and order.** We quote full numeric or rating cells, or full OGL prose sentences, keeping the printed number. We normalise PDF whitespace and line wraps, and leave footnote superscripts out of sentence text. We read in policy workbook row/column order, then PDF page/position, then aggregate workbook table/row/column. In the 28 October file that we cite in rows, the physical page equals the printed page + 3 (+ 4 in the 30 October correction). ## Coverage by measure_key, including null -99 keys on 2,207 rows, 1,577 of them `row_type: measure_component`. Of the 593 null rows, 535 carry a proposed key (374 of them `measure_component`), and 58 are context: 42 fiscal aggregates, 13 headroom rows, Box 3.1's two measure counts and paragraph 3.28's salary-sacrifice revision. Across the family, 1,951 rows are `measure_component`, 649 `measure_total` and 200 carry no row type. The uncertainty ratings are keyed to their measures. +We use 128 keys on 2,742 rows, 1,951 of them `row_type: measure_component`. Of those rows, we key 535 (374 of them `measure_component`) to the 29 keys we proposed and registered in tranche 2. We stage 58 rows as context: 42 fiscal aggregates, 13 headroom rows, Box 3.1's two measure counts and paragraph 3.28's salary-sacrifice revision. Across the family we mark 1,951 rows `measure_component` and 649 `measure_total`, and give 200 no row type. We key the uncertainty ratings to their measures. | measure_key | Rows | |---|---:| @@ -140,52 +150,53 @@ The resolved ZIP URL is the publication URL for both members, with the member na | `ab2026__writing_down_allowance_12pc_and_fya_50pc` | 43 | | `ab2026__your_first_home_equity_loan` | 19 | | `ab2026__youth_guarantee_and_apprenticeships_spring_2026` | 18 | -| `null` | 593 | +| the 29 keys we proposed (below) | 535 | +| `null` (context) | 58 | -## Proposed measure keys and exact producer scopes +## Keys we proposed, with the producer's scopes -All of these rows have `measure_key: null`. Names follow README rule 2 as harmonised on 3 November (review finding 9): a single measure is `ab2026__`, and a scope package's slug names the scope and carries no `obr`, except the OBR's own residual groups, which keep it (4 November, review B finding 15). Where the OBR publishes only a named residual group, its line decomposition stays unresolved; no guessed membership or registry key is attached. +We gave all of these rows `measure_key: null` and proposed these keys, and in tranche 2 we registered them and moved them into `measure_key`. We name them under README rule 2 as harmonised on 3 November (review finding 9). We give a single measure `ab2026__`, and in a scope package's slug we name the scope and drop `obr`. We keep `obr` only for the OBR's own residual groups (4 November, review B finding 15). Where the OBR publishes only a named residual group, we leave its line decomposition unresolved and attach no guessed membership or registry key. -| Proposed key | Rows | Scope | +| Key | Rows | Scope | |---|---:|---| -| `ab2026__departmental_spending_growth_2031_32` | 13 | No Table 6.1 lines. “Departmental spending: Grow day-to-day spending by 0.7% in real terms in 2031-32, matching 0.7% average annual real terms growth from 2029-30 to 2030-31”. PDF describes the same baseline change as a £5.5bn reduction in the 2031-32 pre-measures RDEL assumption. (Was `ab2026__package_obr_departmental_spending_growth_2031_32`.) | -| `ab2026__package_carbon_price_support_and_cfd_levies` | 2 | Table 6.1 lines 70 and 71: row 2673's £0.4bn is their receipts (`measure_component`, excluding the backdated AME support of footnote 9), and row 2677's £0.5bn their totals. The CfD-levy funding sits in line 70: its "Environmental levies" receipts (−480 / −63 / −130 / −408 / −450) are the only environmental-levies effect of any decision (review B finding 7). Footnote 9's £0.8bn of backdated BICS and Supercharger support is line 70's current AME in 2027-28 and is keyed to line 70 as a component. The 28 October PDF dated CPS abolition April 2027; the OBR's 30 October correction moved it to April 2028, matching line 71. (Was `ab2026__package_obr_carbon_price_support_and_cfd_levies`.) | -| `ab2026__package_civil_service_transformation_programme` | 1 | The OBR's “civil service transformation programme ... reduce resource spending”. Membership of Table 6.1 lines 63 and 64 is probable but not established: the OBR names no lines, and lines 63 and 64's totals average £0.56bn from 2028-29 against its £0.5bn (review B finding 14). (Was `ab2026__package_obr_civil_service_transformation_programme`.) | -| `ab2026__package_compliance_measures_announced` | 1 | No Table 6.1 line list. Box 3.1 (p.66): "Of the compliance measures announced in this Budget". The OBR does not list them; they are probably Table 6.1 line 45's items and line 46, but that is not stated, so the scope is the OBR's words. | -| `ab2026__package_corporation_and_bank_tax_changes` | 7 | OBR “Changes to corporation and bank taxation”; paragraph 3.16 explicitly discusses line 58 WDA/FYA, line 55 bank surcharge, and line 54 overseas sovereign immunity. Complete Table 6.1 line decomposition and fiscal-head allocation are not published; unresolved, no inferred additional membership. (Was `ab2026__package_obr_corporation_and_bank_tax_changes`.) | -| `ab2026__package_defence_investment_plan` | 1 | OBR “funding for the defence investment plan” / “the defence investment plan, which increases departmental resource spending (RDEL)”; related to Table 6.1 Defence Investment Plan entries, but complete line decomposition is not published and is unresolved. No inferred40–43 membership. (Was `ab2026__package_obr_defence_investment_plan`.) Shared with the red book's Defence Investment Plan row. Row 2522's £1.8bn a year is the package's RDEL part (`measure_component`, review A's K-2): it matches gross RDEL (line 40's averages £1,766m), not the lines' net effect. | -| `ab2026__package_dividend_and_capital_gains_tax_changes` | 6 | Table 6.1 lines 26,49 and 60: OBR “Changes to dividend and capital gains taxation”. (Was `ab2026__package_obr_dividend_and_capital_gains_tax_changes`.) | -| `ab2026__package_fuel_duty_and_electricity_vat` | 11 | The receipts of Table 6.1 lines 2 and 3 only (Table 3.1's tax groups are receipts groupings, review B finding 3); OBR “Fuel duty freeze and VAT off domestic energy bills”. Excludes household council-tax rebate line 1. (Was `ab2026__package_obr_fuel_duty_and_electricity_vat`.) | -| `ab2026__package_fuel_duty_freezes_since_2010_11` | 1 | OBR cumulative 2010-11 to 2027-28 fuel duty freezes and 5p cut relative to stated RPI uprating; includes Table 6.1 line 3 and earlier freezes since2010-11. Not the cost of the AB2026 measure alone. (Was `ab2026__package_obr_historical_fuel_duty_freezes_and_5p_cut`.) RF's `ab2026_option__fuel_duty_rpi_uprating_since_2011` is the same pair of paths seen from the other side (RPI uprating since January 2011 against the freezes); the two keys stay separate, and each family's NOTES names the other (review A's K-6). | -| `ab2026__package_government_decisions_since_march_2026` | 340 | Table 6.1 lines 1–98, plus the three fiscally neutral measures (PWLB debt relief, Scottish APD devolution, landlord registration fee) and the two non-scorecard policies listed above. OBR “Total effect of Government decisions”, with direct, indirect, receipt, expenditure and fiscal-head projections distinguished in conditions. Includes indirect effects where stated; not the HMT scorecard-only package. (Was `ab2026__package_obr_government_decisions_since_march_2026`.) Shared with the red book's Table 1.3 "Total effect of policy decisions" and RF's restatement. **Parts and wholes (rule 9, 4 November):** the totals (workbook row 549 and Table 3.1's total) are `measure_total` with no `component`; every part is `measure_component`, with `component` the label its primary prints: the workbook's direct and indirect effects, its total of receipts measures ("Total effects of receipts measures", which keeps `revenue_change`) and its 288 lines by head ("Receipts: " or ": "), Table 3.1's direct and indirect effects, and paragraphs 3.2 and 3.5. Never sum a part with the total. The workbook's total of expenditure measures moved to the spending-decisions key and paragraph 3.4's tax line to the tax-decisions key (review B finding 6). | -| `ab2026__package_national_wealth_fund_and_british_business_bank` | 5 | Table 6.1 lines 29 and 30 together, plus their associated financial transactions in Table 6.2; OBR “funding available for the National Wealth Fund and the British Business Bank ... to invest in financial assets”. Paragraph 3.20's effects: the financing requirement (`cash_requirement_change`, restored 4 November), the net debt effect (£3.4bn) and the borrowing effect (£0.5bn) are wholes; the gross debt (£3.9bn) and the additional income (£0.6bn) are their parts (`measure_component`). (Was `ab2026__package_obr_national_wealth_fund_and_british_business_bank`.) | +| `ab2026__departmental_spending_growth_2031_32` | 13 | No Table 6.1 lines. “Departmental spending: Grow day-to-day spending by 0.7% in real terms in 2031-32, matching 0.7% average annual real terms growth from 2029-30 to 2030-31”. In the PDF the OBR describes the same baseline change as a £5.5bn reduction in the 2031-32 pre-measures RDEL assumption. (Was `ab2026__package_obr_departmental_spending_growth_2031_32`.) | +| `ab2026__package_carbon_price_support_and_cfd_levies` | 2 | Table 6.1 lines 70 and 71: we stage row 2675's £0.4bn on their receipts (`measure_component`, excluding the backdated AME support of footnote 9), and row 2679's £0.5bn on their totals. We place the CfD-levy funding in line 70: on that line the OBR prints "Environmental levies" receipts of −480 / −63 / −130 / −408 / −450, the only environmental-levies effect of any decision (review B finding 7). The OBR counts footnote 9's £0.8bn of backdated BICS and Supercharger support as line 70's current AME in 2027-28, and we key it to line 70 as a component. In the 28 October PDF the OBR dated CPS abolition April 2027, and in its 30 October correction it moved the date to April 2028, matching line 71. (Was `ab2026__package_obr_carbon_price_support_and_cfd_levies`.) | +| `ab2026__package_civil_service_transformation_programme` | 1 | The OBR's “civil service transformation programme ... reduce resource spending”. Table 6.1 lines 63 and 64 probably make up the programme, but the OBR names no lines, and lines 63 and 64's totals average £0.56bn from 2028-29 against its £0.5bn (review B finding 14). (Was `ab2026__package_obr_civil_service_transformation_programme`.) | +| `ab2026__package_compliance_measures_announced` | 1 | No Table 6.1 line list. Box 3.1 (p.66): "Of the compliance measures announced in this Budget". The OBR does not list them; they probably cover Table 6.1 line 45's items and line 46, but the OBR doesn't say so, so we take the scope from its words. | +| `ab2026__package_corporation_and_bank_tax_changes` | 7 | OBR “Changes to corporation and bank taxation”; paragraph 3.16 explicitly discusses line 58 WDA/FYA, line 55 bank surcharge, and line 54 overseas sovereign immunity. The OBR publishes no complete Table 6.1 line decomposition or fiscal-head allocation, and we leave it unresolved and infer no further membership. (Was `ab2026__package_obr_corporation_and_bank_tax_changes`.) | +| `ab2026__package_defence_investment_plan` | 1 | OBR “funding for the defence investment plan” / “the defence investment plan, which increases departmental resource spending (RDEL)”; related to Table 6.1 Defence Investment Plan entries, but the OBR publishes no complete line decomposition, and we infer no membership of lines 40–43. (Was `ab2026__package_obr_defence_investment_plan`.) Shared with the red book's Defence Investment Plan row. We stage row 2522's £1.8bn a year as the package's RDEL part (`measure_component`, review A's K-2): it matches gross RDEL (line 40's averages £1,766m), not the lines' net effect. | +| `ab2026__package_dividend_and_capital_gains_tax_changes` | 6 | Table 6.1 lines 26, 49 and 60: OBR “Changes to dividend and capital gains taxation”. (Was `ab2026__package_obr_dividend_and_capital_gains_tax_changes`.) | +| `ab2026__package_fuel_duty_and_electricity_vat` | 11 | The receipts of Table 6.1 lines 2 and 3 only (the OBR groups receipts in Table 3.1's tax groups, review B finding 3); OBR “Fuel duty freeze and VAT off domestic energy bills”. Excludes household council-tax rebate line 1. (Was `ab2026__package_obr_fuel_duty_and_electricity_vat`.) | +| `ab2026__package_fuel_duty_freezes_since_2010_11` | 1 | OBR cumulative 2010-11 to 2027-28 fuel duty freezes and 5p cut relative to stated RPI uprating; includes Table 6.1 line 3 and earlier freezes since 2010-11. Not the cost of the AB2026 measure alone. (Was `ab2026__package_obr_historical_fuel_duty_freezes_and_5p_cut`.) For `ab2026_option__fuel_duty_rpi_uprating_since_2011`, the Resolution Foundation (RF) compares RPI uprating since January 2011 against the freezes, reversing the OBR's comparison. We keep the two keys separate and name each family in the other's NOTES (review A's K-6). | +| `ab2026__package_government_decisions_since_march_2026` | 340 | Table 6.1 lines 1–98, plus the three fiscally neutral measures (PWLB debt relief, Scottish APD devolution, landlord registration fee) and the two non-scorecard policies listed above. OBR “Total effect of Government decisions”, with direct, indirect, receipt, expenditure and fiscal-head projections distinguished in conditions. Includes indirect effects where stated; not the HMT scorecard-only package. (Was `ab2026__package_obr_government_decisions_since_march_2026`.) Shared with the red book's Table 1.3 "Total effect of policy decisions" and RF's restatement. **Parts and wholes (rule 9, 4 November):** we mark the totals (workbook row 549 and Table 3.1's total) `measure_total` with no `component`, and every part `measure_component`, with `component` the label the OBR prints. We count as parts the workbook's direct and indirect effects, its total of receipts measures ("Total effects of receipts measures", on which we keep `revenue_change`) and its 288 lines by head ("Receipts: " or ": "). We also count Table 3.1's direct and indirect effects and paragraphs 3.2 and 3.5. Never sum a part with the total. We moved the workbook's total of expenditure measures to the spending-decisions key and paragraph 3.4's tax line to the tax-decisions key (review B finding 6). | +| `ab2026__package_national_wealth_fund_and_british_business_bank` | 5 | Table 6.1 lines 29 and 30 together, plus their associated financial transactions in Table 6.2; OBR “funding available for the National Wealth Fund and the British Business Bank ... to invest in financial assets”. Paragraph 3.20's effects: we stage the financing requirement (`cash_requirement_change`, restored 4 November), the net debt effect (£3.4bn) and the borrowing effect (£0.5bn) as wholes, and the gross debt (£3.9bn) and the additional income (£0.6bn) as their parts (`measure_component`). (Was `ab2026__package_obr_national_wealth_fund_and_british_business_bank`.) | | `ab2026__package_non_scorecard_decisions` | 12 | No Table 6.1 lines. “Direct effect of non-scorecard decisions”: exactly the departmental spending assumption and public-pension valuation changes above. (Was `ab2026__package_obr_non_scorecard_decisions`.) | -| `ab2026__package_obr_other_ame_spending_decisions` | 6 | OBR Table 3.1 “Other AME spending decisions”; excludes net unfunded pensions valuation, and is net of the council tax rebate shown on its own line (−£2.6bn in 2026-27): Table 6.2 prints "Other spending decisions" at 0.4 in 2026-27, the rebate's 2.6 plus this group's −2.2 (review B finding 9). (Was `ab2026__package_other_ame_spending_decisions` until 4 November, review B finding 15.) Producer does not supply a full Table 6.1 line-number decomposition in PDF. (Was `ab2026__package_obr_other_ame_spending_decisions`.) | -| `ab2026__package_obr_other_departmental_spending_changes` | 7 | The OBR's Table 3.1 group: departmental spending changes excluding the pensions valuation, including the DEL funding for line 1's council tax rebate (the group, 5.4 / 1.2 / 1.5 / −0.1 / −0.2 / −6.0, equals the workbook's DEL total less the pensions valuation's DEL, which includes line 1's −2,627; review B finding 9). It includes defence, capital spending and the £5.5bn 2031-32 pre-measures RDEL reduction, which has no Table 6.1 line. (Was `ab2026__package_other_departmental_spending_changes` until 4 November, review B finding 15.) Producer does not supply a full line-number decomposition in PDF. (Was `ab2026__package_obr_other_departmental_spending_changes`.) | +| `ab2026__package_obr_other_ame_spending_decisions` | 6 | In Table 3.1 the OBR excludes net unfunded pensions valuation from “Other AME spending decisions”. It also nets out the separately shown council tax rebate (−£2.6bn in 2026-27). In Table 6.2 the OBR prints "Other spending decisions" at 0.4 in 2026-27: the rebate's 2.6 plus this group's −2.2 (review B finding 9). (Was `ab2026__package_other_ame_spending_decisions` until 4 November, review B finding 15.) The OBR gives no full Table 6.1 line decomposition in the PDF. (Was `ab2026__package_obr_other_ame_spending_decisions`.) | +| `ab2026__package_obr_other_departmental_spending_changes` | 7 | The OBR's Table 3.1 group: departmental spending changes excluding the pensions valuation, including the DEL funding for line 1's council tax rebate (the group, 5.4 / 1.2 / 1.5 / −0.1 / −0.2 / −6.0, equals the workbook's DEL total less the pensions valuation's DEL, which includes line 1's −2,627; review B finding 9). It includes defence, capital spending and the £5.5bn 2031-32 pre-measures RDEL reduction, which has no Table 6.1 line. (Was `ab2026__package_other_departmental_spending_changes` until 4 November, review B finding 15.) The OBR gives no full Table 6.1 line decomposition in the PDF. (Was `ab2026__package_obr_other_departmental_spending_changes`.) | | `ab2026__package_obr_other_hmrc_compliance_and_debt` | 1 | OBR paragraph 3.13 residual “Other measures”: direct debt recovery, e-invoicing, debt collection agencies (components of Table 6.1 line 45), and online marketplace liability (line 46). (Was `ab2026__package_obr_other_hmrc_compliance_and_debt` until 3 November, and `ab2026__package_other_hmrc_compliance_and_debt` until 4 November.) | | `ab2026__package_obr_other_tax_measures` | 6 | OBR Table 3.1 “Other tax measures”; residual of Table 6.1 tax measures after tax administration/compliance/debt, corporation/bank, savings/property, dividend/CGT and fuel/electricity VAT groups. PDF does not enumerate a full line-number decomposition. (Was `ab2026__package_obr_other_tax_measures` until 3 November, and `ab2026__package_other_tax_measures` until 4 November.) | | `ab2026__package_planned_capital_spending_reductions` | 1 | OBR “Reductions in planned capital spending” / “changes to departmental capital spending”; PDF does not enumerate the exact Table 6.1 lines. (Was `ab2026__package_obr_planned_capital_spending_reductions`.) | -| `ab2026__package_property_and_savings_income_rates_receipts` | 7 | The receipts of Table 6.1 lines 61 and 62: Table 3.1's “Changes to savings and property income tax rates”. (Was `ab2026__package_obr_savings_and_property_income_tax_rates`.) **Receipts only (3 November):** Table 3.1 prints 0.1 in 2027-28 and 0.9 in 2029-30, against lines 61 and 62's totals of 160 and 954 (£m), but in line with their receipts (Table 3.3 gives line 61's as 2, 543 and 400 from 2027-28). So it is not `ab2026__package_property_and_savings_income_rates`, which is exactly lines 61 and 62. | -| `ab2026__package_spending_decisions_since_march_2026` | 13 | OBR “Direct effect of spending measures” / “Spending measures”: expenditure-side effects of decisions since March, including unfunded public service pensions valuation and the £5.5bn 2031-32 RDEL assumption reduction. This is not the HMT total-spending package. Table 6.1 lines 1–98 can affect expenditure fiscal heads; a complete narrower line decomposition is not published and remains unresolved. (Was `ab2026__package_obr_spending_decisions_since_march`.) Since 4 November it also holds the workbook's total of expenditure measures (I520:N520), Table 3.1's direct effect of spending measures reversed (review B finding 6). | -| `ab2026__package_tax_administration_compliance_and_debt_collection_receipts` | 6 | The receipts of Table 6.1 lines 45 and 46: OBR “Tax administration, compliance and debt collection measures”. The group (−0.1 / −0.5 / −0.6 / −2.4 / −2.9 / −3.6) equals lines 45 and 46's receipts (107 / 455 / 567 / 2,411 / 2,881 / 3,630) in all six years; their totals (8 / 262 / 434 / 2,248 / 2,733 / 3,423) round differently in every year (review B finding 3). It must not be attached to line 45 alone. (Was `ab2026__package_obr_tax_administration_compliance_and_debt`, then `ab2026__package_tax_administration_compliance_and_debt_collection` until 4 November.) Shared with RF's restatement. | -| `ab2026__package_tax_decisions_since_march_2026` | 6 | OBR Table 3.1 "Direct effect of tax measures": the direct effect of tax decisions since March 2026, as the OBR classifies them. Shared with RF's restatement of the same row. It equals −1 × the workbook's total of receipts measures, so it is a receipts grouping. Before 3 November these rows sat under the total package with `component: tax measures (OBR classification)`. Since 4 November it also holds paragraph 3.4's "reducing borrowing by an average of £6 billion over the final four years", the mean of this line from 2028-29 (review B finding 6). | -| `ab2026__package_temporary_business_rates_reliefs_receipts` | 1 | The receipts of Table 6.1 lines 12, 13 and 14 (paragraph 3.19's "net cost of around £0.5 billion in 2028-29" is their receipts, 536, against totals of 663; review B finding 3 and review A's K-7(g)): “package of temporary business rates measures” including real-terms freeze for certain sectors, pubs/live-music reliefs, small-business thresholds and extension of transitional support, partly offset by transitional relief supplement. (Was `ab2026__package_obr_temporary_business_rates_reliefs`, then `ab2026__package_temporary_business_rates_reliefs` until 4 November.) | +| `ab2026__package_property_and_savings_income_rates_receipts` | 7 | The receipts of Table 6.1 lines 61 and 62: Table 3.1's “Changes to savings and property income tax rates”. (Was `ab2026__package_obr_savings_and_property_income_tax_rates`.) **Receipts only (3 November):** The OBR prints −0.1 in 2027-28 and −0.9 in 2029-30 in Table 3.1, with positive values for borrowing increases. We match their rounded magnitudes to receipts, rather than to the policy workbook's line totals of £160m and £954m. In Table 3.3 the OBR prints line 61's receipts as −2, −543 and −400 (£m) from 2027-28. So we don't use `ab2026__package_property_and_savings_income_rates`, which covers exactly lines 61 and 62. | +| `ab2026__package_spending_decisions_since_march_2026` | 13 | OBR “Direct effect of spending measures” / “Spending measures”: expenditure-side effects of decisions since March, including unfunded public service pensions valuation and the £5.5bn 2031-32 RDEL assumption reduction. We don't use the HMT total-spending package here. Table 6.1 lines 1–98 can affect expenditure fiscal heads; the OBR publishes no complete narrower line decomposition, and we leave it unresolved. (Was `ab2026__package_obr_spending_decisions_since_march`.) Since 4 November we also key the workbook's total of expenditure measures (I520:N520) here, the reverse of Table 3.1's direct effect of spending measures (review B finding 6). | +| `ab2026__package_tax_administration_compliance_and_debt_collection_receipts` | 6 | The receipts of Table 6.1 lines 45 and 46: OBR “Tax administration, compliance and debt collection measures”. The group (−0.1 / −0.5 / −0.6 / −2.4 / −2.9 / −3.6) equals lines 45 and 46's receipts (107 / 455 / 567 / 2,411 / 2,881 / 3,630) in all six years; their totals (8 / 262 / 434 / 2,248 / 2,733 / 3,423) round differently in every year (review B finding 3). Never key it to line 45 alone. (Was `ab2026__package_obr_tax_administration_compliance_and_debt`, then `ab2026__package_tax_administration_compliance_and_debt_collection` until 4 November.) Shared with RF's restatement. | +| `ab2026__package_tax_decisions_since_march_2026` | 6 | OBR Table 3.1 "Direct effect of tax measures": the direct effect of tax decisions since March 2026, as the OBR classifies them. Shared with RF's restatement of the same row. It equals −1 × the workbook's total of receipts measures, so the OBR groups receipts here. Before 3 November we put these rows under the total package with `component: tax measures (OBR classification)`. Since 4 November we also key paragraph 3.4's "reducing borrowing by an average of £6 billion over the final four years" here, the mean of this line from 2028-29 (review B finding 6). | +| `ab2026__package_temporary_business_rates_reliefs_receipts` | 1 | The receipts of Table 6.1 lines 12, 13 and 14 (paragraph 3.19's "net cost of around £0.5 billion in 2028-29" covers their receipts, 536, against totals of 663; review B finding 3 and review A's K-7(g)): “package of temporary business rates measures” including real-terms freeze for certain sectors, pubs/live-music reliefs, small-business thresholds and extension of transitional support, partly offset by transitional relief supplement. (Was `ab2026__package_obr_temporary_business_rates_reliefs`, then `ab2026__package_temporary_business_rates_reliefs` until 4 November.) | | `ab2026__private_landlord_register_fee_65` | 18 | No Table 6.1 lines. “Landlord registration fee”; receipts, current DEL and total. (Was `ab2026__package_obr_landlord_registration_fee`.) The costings family's spelling for the same measure. | -| `ab2026__public_sector_pensions_2024_valuations` | 27 | No Table 6.1 lines. “Public Sector Pensions: Net impact of the 2024 valuations”; departmental reductions, net public service pension payments and their net total. PDF calls this “Net impact of unfunded pensions valuation”. (Was `ab2026__package_obr_public_sector_pensions_2024_valuations`.) | +| `ab2026__public_sector_pensions_2024_valuations` | 27 | No Table 6.1 lines. “Public Sector Pensions: Net impact of the 2024 valuations”; departmental reductions, net public service pension payments and their net total. In the PDF the OBR calls this “Net impact of unfunded pensions valuation”. (Was `ab2026__package_obr_public_sector_pensions_2024_valuations`.) | | `ab2026__pwlb_debt_relief_councils_severe_distress` | 12 | No Table 6.1 lines. “Provide Public Works Loan Board (PWLB) debt relief for councils in severe financial distress”; capital AME and total. (Was `ab2026__package_obr_pwlb_distress_debt_relief`.) The costings family's spelling for the same measure. | | `ab2026__scottish_air_passenger_duty_devolution` | 18 | No Table 6.1 lines. “Devolving Air Passenger Duty to the Scottish Government in April 2027”; Scottish BGA, self-financed expenditure and total. (Was `ab2026__package_obr_scottish_air_passenger_duty_devolution`.) | -| `ab2026__state_pension_triple_lock_adjustment` | 2 | No Table 6.1 line: OBR paragraph 3.49 adjusted state pension triple lock expected April 2030, retaining CPI/2.5percent floor but linking catch-up to2029-30 earnings ratio. Figures 3.51 are historical simulations1990–2025, not the medium-term policy costing. (Was `ab2026__package_obr_adjusted_state_pension_triple_lock`.) The red book's spelling for the same measure. | -| `ab2026_option__fuel_duty_continued_freeze` | 1 | No Table 6.1 line as an announced measure: OBR paragraph 3.43 counterfactual “if fuel duty were to remain frozen until 2029-30”, extending line 3 beyond July2027. (Was `ab2026_option__obr_fuel_duty_frozen_until_2029_30`.) Shared with RF, whose series for the OBR's constant-duty-rate alternative prints the same £4.3bn in 2029-30. | -| `ab2026_option__health_foundation_social_care_reforms` | 2 | No Table 6.1 line: Health Foundation illustrative future social care reforms cited by OBR paragraph 3.41, from limited home-care reform to wider schemes including residential care; costs additional to current spending in 2035-36 per footnote 13. This is not a settled AB2026 policy. (Was `ab2026_option__obr_health_foundation_social_care_reforms`.) | +| `ab2026__state_pension_triple_lock_adjustment` | 2 | No Table 6.1 line: OBR paragraph 3.49 adjusted state pension triple lock expected April 2030, retaining CPI/2.5 percent floor but linking catch-up to 2029-30 earnings ratio. Paragraph 3.51's two figures (rows 2743–2744) come from the OBR's simulation over 1990–2025, not from a medium-term costing of the policy. (Was `ab2026__package_obr_adjusted_state_pension_triple_lock`.) The red book's spelling for the same measure. | +| `ab2026_option__fuel_duty_continued_freeze` | 1 | No Table 6.1 line as an announced measure: OBR paragraph 3.43 counterfactual “if fuel duty were to remain frozen until 2029-30”, extending line 3 beyond July 2027. (Was `ab2026_option__obr_fuel_duty_frozen_until_2029_30`.) Shared with RF, which prints the same £4.3bn in 2029-30 in its series for the OBR's constant-duty-rate alternative. | +| `ab2026_option__health_foundation_social_care_reforms` | 2 | No Table 6.1 line: Health Foundation illustrative future social care reforms cited by OBR paragraph 3.41, from limited home-care reform to wider schemes including residential care; costs additional to current spending in 2035-36 per footnote 13. The OBR cites the Health Foundation's estimates as illustrations, and HM Treasury prints no social care line in Table 6.1. (Was `ab2026_option__obr_health_foundation_social_care_reforms`.) | ## Metric, unit and baseline proposals -No unregistered enum literals are used. `fiscal_headroom` is registered in the live models even though the older README calls it a proposal. +We use no unregistered enum literals. We register `fiscal_headroom` in the live models, even though we call it a proposal in the older README. | proposed_metric | Rows | Scope | |---|---:|---| -| `measure_count` | 2 | Box 3.1: the 15 measures expected to reduce the tax gap, and the four of them that target the small business tax gap (context counts, as the uk_thinktank_misc and uk_ifs_ab2026 families stage measure counts; review B finding 11). | +| `measure_count` | 2 | Box 3.1: the 15 measures expected to reduce the tax gap, and the four of them that target the small business tax gap (context counts, as we stage measure counts in the uk_thinktank_misc and uk_ifs_ab2026 families; review B finding 11). | | `cash_receipts_change` | 1 | Self-assessment overlapping transition payments, £17bn in 2029-30; assumed extraordinary accruals adjustment excludes the temporary cash effect from PSNB. | | `fiscal_balance_change` | 6 | Table 3.1's memo line, the total effect of Government decisions on the current budget deficit (README rule 10, 3 November). | | `tax_gap_share_of_compliance_yield` | 1 | Box 3.1: the share of this Budget's compliance measures' yield the OBR expects to reduce the tax gap ("a third"). | @@ -198,13 +209,13 @@ No unregistered enum literals are used. `fiscal_headroom` is registered in the l | `public_sector_net_debt_change` | 3 | Marginal stock debt effects of NWF/BBB investment and Your First Home lending, paragraphs 3.20/3.22. | | `public_sector_net_financial_liabilities` | 7 | End-March PSNFL stock, GBP. | -Proposed units: `percent_of_gdp` (28 rows, nominal GDP denominator, including one prose package effect and fiscal shares/headroom); `rating_category` (7 High uncertainty ratings); `measures` (Box 3.1's 2 counts). +We propose these units: `percent_of_gdp` (28 rows, nominal GDP denominator, including one prose package effect and fiscal shares/headroom); `rating_category` (7 High uncertainty ratings); `measures` (Box 3.1's 2 counts). -Registered metrics beyond `revenue_change`: `decisions_effect_on_borrowing` (91 rows, README rule 10), `fiscal_headroom` (13), `cpi_inflation_effect` (6), `benefit_cost_change` (2, paragraph 3.51's simulations), `cash_requirement_change` (1, paragraph 3.20's financing requirement) and `gdp_level_effect` (1). +Beyond `revenue_change`, we use these registered metrics: `decisions_effect_on_borrowing` (91 rows, README rule 10), `fiscal_headroom` (13), `cpi_inflation_effect` (6), `benefit_cost_change` (2, paragraph 3.51's simulations), `cash_requirement_change` (1, paragraph 3.20's financing requirement) and `gdp_level_effect` (1). -**For tranche 2:** HM Treasury's Table 6.2 (in `uk_hmt_tables`) stages the cash impacts of the same NWF and BBB financial transactions on the proposed `financial_transaction_cash_impact`, while paragraph 3.20's £1.9bn uses the registered `cash_requirement_change`. Both measure the cash effect of the same transactions, one as a cash impact and one as the financing requirement; the ingest settles the metric. +**At ingest.** In `uk_hmt_tables` we stage HM Treasury's Table 6.2 cash impacts of the same NWF and BBB financial transactions on the proposed `financial_transaction_cash_impact`, and here we stage paragraph 3.20's £1.9bn (row 2684) on the registered `cash_requirement_change`. At ingest we put Table 6.2's 60 rows on `cash_requirement_change` too, so both producers' figures share one metric. -Every `baseline_policy` is null; effect estimates use the following explicit unregistered baseline proposals. Rating, funding-level and aggregate context rows have no proposed baseline. +We leave every `baseline_policy` null, and give effect estimates the explicit unregistered baseline proposals below. We propose no baseline for rating, funding-level and aggregate context rows. | proposed_baseline, verbatim | Rows | |---|---:| @@ -217,106 +228,117 @@ Every `baseline_policy` is null; effect estimates use the following explicit unr ## Not staged and unreachable material -- **Design (rule 8), dropped on 3 November (review finding 1), 57 rows (58 until paragraph 3.20's financing requirement was restored on 4 November):** - - Table 3.2's 47 funding cells: HMRC and DWP baseline, Budget 2025, Budget 2026 additional and total compliance, debt and fraud-and-error funding ("the funding provision", Source: HM Treasury). The table prints no yield, and its Budget 2026 lines are not the scored costs (its DWP line prints −16, 188, 178, 188, 210, 214; the workbook's line-47 departmental spending prints 218, −194, −184, −195, −217, −221). - - Six staffing inputs: the 2,000, 200 and 150 DWP staff (para 3.12); the 5,500 and 670 HMRC staff and the "23 per cent increase" (para 3.13). They now sit in `conditions.staffing` on the £0.5bn and £0.4bn yield rows. +- **Design (rule 8), which we dropped on 3 November after review finding 1: 57 rows (58 until we restored paragraph 3.20's financing requirement on 4 November).** + - Table 3.2's 47 funding cells: HMRC and DWP baseline, Budget 2025, Budget 2026 additional and total compliance, debt and fraud-and-error funding ("the funding provision", Source: HM Treasury). The OBR prints no yield in the table, and its Budget 2026 lines differ from the scored costs. For DWP the OBR prints −16, 188, 178, 188, 210, 214 in Table 3.2 and 218, −194, −184, −195, −217, −221 on the workbook's line-47 departmental spending row. + - Six staffing inputs: the 2,000, 200 and 150 DWP staff (para 3.12); the 5,500 and 670 HMRC staff and the "23 per cent increase" (para 3.13). We now keep them in `conditions.staffing` on the £0.5bn and £0.4bn yield rows. - Guarantee capacity, a contingent liability: UK Export Finance's £40bn and the Growth Guarantee Scheme's £7bn (para 3.21). - - A lending allocation: Your First Home's £0.7bn and £1bn of loans (para 3.22). Paragraph 3.22's effect on debt (£0.7bn in the first year) is staged. -- **The rule for lending and financing (4 November, the coordinator's call on review B's finding 5; README rule 8):** the harvest stages HM Treasury's Tables 6.1, 6.2 and 6.4 and the OBR's policy workbook whole, as the decisions' costings. Elsewhere, an amount a scheme will lend or spend is design (rule 8), and an effect on borrowing, debt or the financing requirement is an estimate. So paragraph 3.22's loan volumes stay dropped, its effect on debt stays, and paragraph 3.20's "£1.9 billion a year" increase in the Government's financing requirement, dropped on 3 November as a financing allocation, is restored as a fiscal-aggregate estimate (`cash_requirement_change`). EFO paragraphs 6.19–6.20 say the CGNCR "forms the basis of" the DMO's remit, so the requirement is the aggregate and the remit is the design. -- **Repeats, dropped on 3 November (review finding 5, the coordinator's call), 37 rows.** The table cells are the staged rows, as in the AB2025 OBR family: text that repeats a Table 3.1 cell (paras 3.2, 3.3, 3.4, 3.11, 3.13, 3.16, 3.17 and 3.19: 19 rows) or earlier text (paras 3.11, 3.12, 3.13 and 3.16: 6 rows), and Table 3.1's "Spending measures" and "Tax measures" rows, which repeat its "Direct effect of spending measures" and "Direct effect of tax measures" rows (12 rows). - -- **Chapters 1, 2 and 4–6 and paragraph 7.20 (4 November, review B finding 4; the coordinator's call: not staged).** Most of these chapters repeat Table 3.1 and the workbook's lines by head (Tables 4.5–4.7, 5.4, 5.6 and 6.2 among them). These estimates are new and are not staged; tranche 2 may add any of them as components: - - Tables 4.2 (p.87, indirect effects on receipts), 5.2 (p.108, indirect effects on spending) and 5.8 (p.128, "Effects of Government decisions" on debt interest, −0.9 / 1.2 / 1.0 / 0.7 / 0.4 / 0.3): parts of the decisions' total effect, which rows 2490–2495 and 2534–2539 (4f424b9 order) stage whole. - - Tables 4.3 to 4.7's indirect rows by tax and their PAYE / self-assessment split, including Table 4.4's self-assessment income tax by measure (e.g. "Timely payment for self assessment" 3.5 / 6.5 in 2029-30 / 2030-31): splits of lines the family already stages, by tax or by cash timing. - - Paragraph 4.30: a continued fuel duty freeze "would reduce receipts, on average, by £4 billion a year between 2027-28 and 2031-32", the receipts average of the option whose effect on borrowing row 2737 stages. - - Table 6.3 (p.141, "Effects of policy on financial transactions", −0.1 / 0.8 / 2.1 / −12.4 / 3.9 / 4.0): the OBR's whole-package effect on financial transactions; tranche 2 can stage it beside HM Treasury's Table 6.2 total, which `uk_hmt_tables` stages. + - A lending allocation: Your First Home's £0.7bn and £1bn of loans (para 3.22). We stage paragraph 3.22's effect on debt (£0.7bn in the first year). +- **The rule for lending and financing (4 November, the coordinator's call on review B's finding 5; README rule 8).** In the harvest we stage HM Treasury's Tables 6.1, 6.2 and 6.4 and the OBR's policy workbook whole, as the decisions' costings. Elsewhere we treat an amount a scheme will lend or spend as design (rule 8), and an effect on borrowing, debt or the financing requirement as an estimate. So we keep paragraph 3.22's loan volumes dropped and its effect on debt staged. We had dropped paragraph 3.20's "£1.9 billion a year" increase in the Government's financing requirement on 3 November as a financing allocation, and we restored it as a fiscal-aggregate estimate (`cash_requirement_change`). In EFO paragraphs 6.19–6.20 the OBR says the CGNCR "forms the basis of" the DMO's remit, so we treat the requirement as the aggregate and the remit as the design. +- **Repeats, which we dropped on 3 November (review finding 5, the coordinator's call): 37 rows.** As in the AB2025 OBR family, we stage the table cells. We dropped text that repeats a Table 3.1 cell (paras 3.2, 3.3, 3.4, 3.11, 3.13, 3.16, 3.17 and 3.19: 19 rows) or earlier text (paras 3.11, 3.12, 3.13 and 3.16: 6 rows). We also dropped Table 3.1's "Spending measures" and "Tax measures" rows, which repeat its "Direct effect of spending measures" and "Direct effect of tax measures" rows (12 rows). + +- **Chapters 1, 2 and 4–6 and paragraph 7.20 (4 November, review B finding 4; the coordinator's call: not staged).** In most of these chapters the OBR repeats Table 3.1 and the workbook's lines by head (Tables 4.5–4.7, 5.4, 5.6 and 6.2 among them). The OBR gives these new estimates there, which we didn't stage; tranche 2 may add any of them as components: + - Tables 4.2 (p.87, indirect effects on receipts), 5.2 (p.108, indirect effects on spending) and 5.8 (p.128, "Effects of Government decisions" on debt interest, −0.9 / 1.2 / 1.0 / 0.7 / 0.4 / 0.3). They form parts of the decisions' total effect, which we stage whole in rows 2490–2495 and 2534–2539 (4f424b9 order). + - Tables 4.3 to 4.7's indirect rows by tax and their PAYE / self-assessment split, including Table 4.4's self-assessment income tax by measure (e.g. "Timely payment for self assessment" 3.5 / 6.5 in 2029-30 / 2030-31). They split, by tax or by cash timing, lines we already stage. + - Paragraph 4.30: a continued fuel duty freeze "would reduce receipts, on average, by £4 billion a year between 2027-28 and 2031-32". The OBR gives here the receipts average of the option whose effect on borrowing we stage in row 2740. + - Table 6.3 (p.141, "Effects of policy on financial transactions", −0.1 / 0.8 / 2.1 / −12.4 / 3.9 / 4.0): the OBR's whole-package effect on financial transactions. We don't stage it yet. We stage HM Treasury's Table 6.2 total in `uk_hmt_tables`. - Paragraph 7.20's decomposition of the changes in the fiscal-target margins. -- AnnexA has no policy-costing narrative beyond introductionA.1; current receipts/spending/economy forecast TablesA.1–A.8/A.11–A.12 are outside the policy-family scope. Headline fiscal aggregates inA.9 are owned by aggregate lane. -- No printed numeric Pension Tax Allowance or Carer’s Allowance taper claim appears in chapter3 or AnnexA. Whole-PDF search finds no PTA phrase; Carer’s Allowance appears only in a generic welfare footnote, not taper costing. These measures are covered by workbook rows. -- Charts3.1–3.4 and Box3.1 ChartsA/B contain no printed data labels for the policy values; no bars or paths read. Tick labels and dates are not claims. Box 3.1's tax-gap context restates HMRC and several-Budget historical compliance yields and is not staged. Its three AB2026 figures are staged: the estimate "a third of the yield" (review finding 10, the sentence completed on 4 November) and the two measure counts, 15 measures expected to reduce the tax gap and four that target the small business tax gap (review B finding 11). -- Policy design parameters (tax rates/thresholds, dates, litres/pence) without costing or behavioural-estimate content were not separately staged. -- Older-policy employment/housing/childcare monitoring examples in 3.29–3.36 do not cost an AB2026 Table 6.1 measure; no new AB2026 measure key invented. Paragraph 3.28’s salary-sacrifice figure is the OBR's +£0.4bn revision to its Autumn Budget 2025 costing: a context row with a null key since 4 November, since a revision is not a level and doesn't share a key with Table 6.4's recosted level (review A's H-6(b), the coordinator's call). -- Paragraph3.48 FRS long-run pension/debt scenarios are current-policy/earlier-publication context, outside requested headline medium-term aggregate scope. Paragraph3.52 adjusted-triple-lock debt effects are explicitly2075-76=>2076; checker caps periods at2075, so neither25pp nor10pp long-run debt effect was staged or misdated. -- No additional primary access attempts owned by text lane; root records direct/local recipe and unreachable EFO landing page in manifest/NOTES. +- In Annex A the OBR gives no policy-costing narrative beyond introduction A.1. Its current receipts, spending and economy forecast Tables A.1–A.8/A.11–A.12 fall outside the policy family's scope. The aggregate lane owns the headline fiscal aggregates in A.9. +- The OBR prints no numeric Pension Tax Allowance or Carer’s Allowance taper claim in chapter 3 or Annex A. A whole-PDF search finds no PTA phrase, and Carer’s Allowance appears only in a generic welfare footnote, not a taper costing. We cover these measures with workbook rows. +- The OBR prints no data labels for the policy values in Charts 3.1–3.4 and Box 3.1 Charts A/B, and we read no bars or paths. We don't count tick labels and dates as claims. In Box 3.1's tax-gap context the OBR restates HMRC and several Budgets' historical compliance yields, which we didn't stage. We stage its three AB2026 figures. In one the OBR estimates "a third of the yield" (review finding 10; we completed the sentence on 4 November). In the other two it counts measures: "15 measures expected to reduce the tax gap", and four that target the small business tax gap (review B finding 11). +- We didn't separately stage policy design parameters (tax rates/thresholds, dates, litres/pence) in which the OBR gives no costing or behavioural estimate. +- In paragraphs 3.29–3.36 the OBR monitors older employment, housing and childcare policies and costs no AB2026 Table 6.1 measure there, so we invented no new AB2026 measure key for them. In paragraph 3.28 the OBR revises its Autumn Budget 2025 costing of salary sacrifice by +£0.4bn. Since 4 November we stage that as a context row with a null key (review A's H-6(b), the coordinator's call). A revision isn't a level, so it doesn't share a key with Table 6.4's recosted level. +- In paragraph 3.48 the OBR gives FRS long-run pension and debt scenarios as current-policy or earlier-publication context, outside the requested headline medium-term aggregate scope. In paragraph 3.52 the OBR dates the adjusted-triple-lock debt effects 2075-76, which would map to period 2076. The checker caps periods at 2075, so we staged neither the 25pp nor the 10pp long-run debt effect, rather than misdate them. +- The text lane owns no further primary access attempts; in the root manifest and NOTES we record the direct/local recipe and the unreachable EFO landing page. -- The policy workbook contains no static/post-behavioural split or rating matrix, and no recosting table. Fiscal-head rows are not labeled as static or behavioural. Contents/navigation cells are not additional claims. -- The supplied chapter-2 charts workbook and the other economy, receipts, expenditure and debt-interest workbooks are outside the policy/headline scope. The chapter-2 fresh URL was 403; it was not used. No economy-wide GDP/CPI paths were staged; only explicit policy CPI/GDP effects. -- The EFO landing page, resolved detailed-tables ZIP and chapter-2 workbook URLs were unreached (403). The PDF attachment URL could not be discovered. No missing online content, new attachments or separate latest ratings database was presumed available or silently substituted. The seven printed PDF ratings are included. +- The OBR gives no static/post-behavioural split, rating matrix or recosting table in the policy workbook, and labels no fiscal-head rows as static or behavioural. We don't count contents or navigation cells as further claims. +- The supplied chapter-2 charts workbook and the other economy, receipts, expenditure and debt-interest workbooks fall outside the policy/headline scope. The chapter-2 fresh URL returned 403, and we didn't use it. We staged no economy-wide GDP/CPI paths, only explicit policy CPI/GDP effects. +- We couldn't reach the EFO landing page, resolved detailed-tables ZIP or chapter-2 workbook URLs (403), and couldn't discover the PDF attachment URL. We presumed no missing online content, new attachments or separate latest ratings database available, and substituted none of them. We include the seven printed PDF ratings. ## Contradictions and limitations retained -- Workbook row 549 “Total effect of Government decisions” displays **-8,905 / -50 / 1,322 / 5,484 / 6,425 / 14,093 £m**, while PDF Table 3.1 prints **6.3 / 7.0 / 2.1 / -2.6 / -3.5 / -9.3 £bn**. This exceeds a sign flip or rounding difference. Workbook direct row 551 follows its negative-loss legend; indirect row 555 has the same signs as the PDF borrowing effects. The workbook appears to mix conventions in its total. Both sources remain unchanged; row notes call out the inconsistency. -- Workbook/HMT precision differs: Household Energy Payment is -2,627 £m here versus -2,625 £m in HMT Table 6.1. Rounded components need not sum to the separately rounded totals. -- The 28 October text of paragraph 3.19 dated carbon-price-support abolition April 2027, and the OBR's 30 October correction moved it to April 2028, matching line 71. Its combined CPS/CfD scope (lines 70 and 71) uses a proposal, not the CPS key. -- Paragraph3.16 dates first-year-allowance increase April 2027; HMT line 58 dates1January2027. The review found the date difference a description difference, not a scope difference, so the WDA/FYA row takes line 58's key. -- Paragraph3.21 says£7bn Growth Guarantee expansion; HMT line 75 says£6.5bn additional SME lending. The £7bn and £6.5bn are different quantities of the same measure (guarantee capacity against lending supported); the borrowing rows take line 75's key and the £7bn is not staged. -- Table 3.3/paragraph 3.24 call reform Company Share Option Plan (CSOP); HMT line 28 says Enterprise Management Incentives2 (EMI2). The workbook's line 28 total (−26 / −33 / −47 / −144 / −142) matches Table 3.3, so the review re-keyed it to line 28 as its tax head. -- Paragraph 3.11's "increase borrowing by £5 billion in 2026-27" is Table 3.1's "Other departmental spending changes" (5.4), rounded; that group includes the DEL funding for the council tax rebate, and Table 3.1 nets the rebate out of its "Other AME spending decisions" group instead (review B finding 9). So the rebate row (2.6), the departmental group and the AME group are not additive in 2026-27. -- OBR tax aggregates include fiscal-head/classification differences from HMT:2031-32 OBR -£7.5bn against HMT+£6.970bn. No total-tax registry key used. -- Paragraph3.2 prose rounds2031-32 borrowing reduction to£12bn; Table 3.1 prints-11.7bn. Both preserved. -- Paragraph3.41 says social-care options; footnote 13 calls cited amounts “extra costs of healthcare”. Reported as printed, no correction. -- Health Foundation range is across different reform options, not statistical uncertainty about one fixed policy. -- Whole-Scheme UKEF exports in paragraph 3.21 are not proven identical to defence-export capacity in HMT line 44. India-UK FTA GDP claim does not include Double Contributions Convention component in HMT line 77. The UKEF borrowing row matches line 44 and takes its key; the India FTA GDP row is a component of line 77. +- On workbook row 549 “Total effect of Government decisions” the OBR displays **-8,905 / -50 / 1,322 / 5,484 / 6,425 / 14,093 £m**, while in PDF Table 3.1 it prints **6.3 / 7.0 / 2.1 / -2.6 / -3.5 / -9.3 £bn**. The gap exceeds a sign flip or rounding difference. On workbook direct row 551 the OBR follows its negative-loss legend, and on indirect row 555 it uses the same signs as the PDF borrowing effects. The OBR appears to mix conventions in the workbook's total. We leave both sources unchanged and call out the inconsistency in row notes. +- The OBR and HMT round differently: the OBR gives the Household Energy Payment as -2,627 £m here, and HMT gives -2,625 £m in Table 6.1. Rounded components need not sum to the separately rounded totals. +- In the 28 October text of paragraph 3.19 the OBR dated carbon-price-support abolition April 2027, and in its 30 October correction moved it to April 2028, matching line 71. For its combined CPS/CfD scope (lines 70 and 71) we use our own package key, not the CPS key. +- In paragraph 3.16 the OBR dates the first-year-allowance increase April 2027, while HMT dates line 58 1 January 2027. The reviewer judged the dates a difference of description, not scope, so we key the WDA/FYA row to line 58. +- In paragraph 3.21 the OBR gives a £7bn Growth Guarantee expansion, while HMT gives £6.5bn of additional SME lending on line 75. The £7bn and £6.5bn measure different quantities of the same measure (guarantee capacity against lending supported). We key the borrowing rows to line 75 and don't stage the £7bn. +- In Table 3.3 and paragraph 3.24 the OBR calls the reform Company Share Option Plan (CSOP), while on line 28 HMT says Enterprise Management Incentives 2 (EMI2). The workbook's line 28 total (−26 / −33 / −47 / −144 / −142) matches Table 3.3, so after review we re-keyed it to line 28 as its tax head. +- In paragraph 3.11 the OBR rounds Table 3.1's "Other departmental spending changes" (5.4) to "increase borrowing by £5 billion in 2026-27". In that group it includes the DEL funding for the council tax rebate, and in Table 3.1 the OBR nets the rebate out of its "Other AME spending decisions" group instead (review B finding 9). So don't add the rebate row (2.6), the departmental group and the AME group together in 2026-27. +- The OBR's tax aggregates differ from HMT's in fiscal head and classification: for 2031-32 the OBR gives -£7.5bn against HMT's +£6.970bn. We use no total-tax registry key. +- In paragraph 3.2 the OBR rounds the 2031-32 borrowing reduction to £12bn, and it prints -11.7bn in Table 3.1. We keep both. +- In paragraph 3.41 the OBR speaks of social-care options, while in footnote 13 it calls the cited amounts “extra costs of healthcare”. We report them as printed, with no correction. +- The Health Foundation's range spans different reform options, not statistical uncertainty about one fixed policy. +- In the EFO the OBR never shows that the whole-scheme UKEF exports in paragraph 3.21 match the defence-export capacity in HMT line 44. The OBR's India-UK FTA GDP claim leaves out the Double Contributions Convention component in HMT line 77. We key the UKEF borrowing row, which matches line 44, to that line, and stage the India FTA GDP row as a component of line 77. ## Validation -- **4 November:** the contract checker reports 2,800 rows and 0 problems, and a re-run of `stage.py` in a scratch directory reproduces the gzip byte for byte. No primary was re-opened in this session; the new quotes (Box 3.1's sentences) are the coordinator's reading of its local copy of the 28 October PDF, and the corrected PDF's hash and change history are review B's. -- **3 November:** the contract checker reports 2,797 rows, 0 problems, and a re-run of `stage.py` reproduces the gzip byte for byte (COUNTS.json pins its sha256). No primary was re-opened: the session's network policy refused obr.uk, and this lane's primaries are the coordinator's local copies. The fixes relabel, re-key or drop rows; no printed value changed. -- **29 October:** the lane ran the full-family checker, a deterministic gzip digest and 40 seeded random row checks, plus ten aggregate checks. Independent audits verified all 2,520 workbook cells and all 316 chapter-3 rows. Its REPORT.md, the only report file any family carried, was deleted under review finding 7. +- On 4 November, the contract checker reported 2,800 rows and 0 problems, and when we re-ran `stage.py` in a scratch directory it reproduced the gzip byte for byte. We re-opened no primary in this session. The coordinator read the new quotes (Box 3.1's sentences) from its local copy of the 28 October PDF, and review B supplied the corrected PDF's hash and change history. +- On 3 November, the contract checker reported 2,797 rows, 0 problems, and when we re-ran `stage.py` it reproduced the gzip byte for byte (COUNTS.json pins its sha256). We re-opened no primary: the session's network policy refused obr.uk, and this lane's primaries come from the coordinator's local copies. We only relabelled, re-keyed or dropped rows, and changed no printed value. +- On 29 October, the lane ran the full-family checker, a deterministic gzip digest and 40 seeded random row checks, plus ten aggregate checks. Independent auditors verified all 2,520 workbook cells and all 316 chapter-3 rows. We deleted the lane's REPORT.md, the only report file any family included, under review finding 7. ## Review fixes, 3 November 2026 -The OBR review (`harvest-staging/ab2026/reviews/verify-obr.md` on the staging branch, verdict REWORK) found the numbers sound and the keying, typing and design rows systematic. Every fix is applied in `review_fixes()` in `stage.py`, with the coordinator's calls on findings 5, 6, 9 and 10: - -1. **Design rows (rule 8):** 58 rows dropped (see "Not staged"); the staffing figures moved into the yield rows' conditions. -2. **Components (rule 9):** 37 rows re-keyed to their measure. Direct recovery of debt, e-invoicing mandation, timely payment for self assessment and HMRC's compliance workforce are line 45 components; verification assurance officers are line 47's; the India FTA GDP effect is line 77's; Table 3.3's CSOP reform is line 28's tax head; UKEF's borrowing effect is line 44 (moved to 2030-31, as line 44 is 0 in 2031-32); the Growth Guarantee Scheme's borrowing effects are line 75; the WDA and FYA yield is line 58. Rows 2693 and 2747 were dropped under finding 5, so their re-keys fell away. Nine proposed keys went with these re-keys, and the WDA/FYA one too. -3. **Row types:** Table 3.3's tax-head rows for lines 54, 61 and 46, the three behavioural-erosion rows and paragraph 3.12's welfare saving are `row_type: measure_component`; the bank surcharge's erosion row is `cumulative`. -4. **Sign conventions:** the workbook's indirect-effect rows say they follow PSNB; its total rows say they mix two conventions. -5. **Repeats:** 37 rows dropped (the coordinator's call; see "Not staged"). -6. **Field fixes:** row 2774 uses `decisions_effect_on_borrowing` with `component: gross additional investment income`; row 2828 carries `ab2025__salary_sacrifice_pension_nics_cap_2000` with `recosting_of: autumn_budget_2025`, which the contract accepted. (Overridden on 4 November: row 2734 in the 4f424b9 order is now a context row with a null key; see the end.) -7. **Files and paths:** REPORT.md is deleted; the manifest notes and these notes name no local machine paths; the PDF's manifest note says its origin and hash are recorded only by this lane. -8. **NOTES.md:** tallies, keys and the not-staged list are regenerated from the rows. -9. **Key names:** renamed under rule 2, as the proposed-key table records. The Table 3.1 savings and property grouping is receipts-only, so it has its own key rather than the lines 61 and 62 package. -10. **Box 3.1:** "a third of the yield" is staged from the review's quotation, at medium confidence, under `ab2026__package_compliance_measures_announced`. - -Row indices above are the 29 October order. 2,797 rows: 2,891 − 58 − 37 + 1. +In its review (`harvest-staging/ab2026/reviews/verify-obr.md` on the staging branch, verdict REWORK), the OBR reviewer found the numbers sound and the keying, typing and design rows systematic. We applied every fix in `review_fixes()` in `stage.py`, with the coordinator's calls on findings 5, 6, 9 and 10: + +1. **Design rows (rule 8):** we dropped 58 rows (see "Not staged") and moved the staffing figures into the yield rows' conditions. +2. **Components (rule 9):** we re-keyed 37 rows to their measure. + - Direct recovery of debt, e-invoicing mandation, timely payment for self assessment and HMRC's compliance workforce became line 45 components, and verification assurance officers line 47's. + - The India FTA GDP effect became line 77's, and Table 3.3's CSOP reform line 28's tax head. + - UKEF's borrowing effect went to line 44, moved to 2030-31 because HMT prints 0 on line 44 in 2031-32. The Growth Guarantee Scheme's borrowing effects went to line 75, and the WDA and FYA yield to line 58. + - We dropped rows 2693 and 2747 under finding 5, so their re-keys fell away. We retired nine proposed keys with these re-keys, and the WDA/FYA one too. +3. **Row types:** we marked Table 3.3's tax-head rows for lines 54, 61 and 46, the three behavioural-erosion rows and paragraph 3.12's welfare saving `row_type: measure_component`, and the bank surcharge's erosion row `cumulative`. +4. **Sign conventions:** on the workbook's indirect-effect rows we say they follow PSNB, and on its total rows that they mix two conventions. +5. **Repeats:** we dropped 37 rows (the coordinator's call; see "Not staged"). +6. **Field fixes:** we gave row 2774 `decisions_effect_on_borrowing` with `component: gross additional investment income`. We keyed row 2828 to `ab2025__salary_sacrifice_pension_nics_cap_2000` with `recosting_of: autumn_budget_2025`, which the contract accepted. (We overrode that on 4 November: row 2734 in the 4f424b9 order is now a context row with a null key; see the end.) +7. **Files and paths:** we deleted REPORT.md, removed local machine paths from the manifest notes and these notes, and say in the PDF's manifest note that only this lane recorded its origin and hash. +8. **NOTES.md:** we regenerated the tallies, keys and the not-staged list from the rows. +9. **Key names:** we renamed keys under rule 2, and record them in the key table above. The Table 3.1 savings and property grouping covers receipts only, so we gave it its own key rather than the lines 61 and 62 package. +10. **Box 3.1:** we staged "a third of the yield" from the review's quotation, at medium confidence, under `ab2026__package_compliance_measures_announced`. + +We give row indices above in the 29 October order. 2,797 rows: 2,891 − 58 − 37 + 1. ## README rule 10, 3 and 4 November 2026 -The IFS review of d5e323dc (finding 3) found this family's totals of the Budget's decisions on `revenue_change`. `rule10()` in `stage.py` applies README rule 10 after `review_fixes()`, and moves 81 rows, the OBR's stated effects of the Budget's decisions on PSNB, to `decisions_effect_on_borrowing`: +The IFS reviewer of d5e323dc (finding 3) found this family's totals of the Budget's decisions on `revenue_change`. `rule10()` in `stage.py` applies README rule 10 after `review_fixes()`. It moves 81 rows, the OBR's stated effects of the Budget's decisions on PSNB, to `decisions_effect_on_borrowing`: -- Table 3.1 (42 rows): the total, direct and indirect effects of Government decisions, the direct effects of tax and of spending measures, and the two memo lines that split the direct effect between decisions on and off HM Treasury's policy decisions table. +- Table 3.1 (42 rows): the total, direct and indirect effects of Government decisions, and the direct effects of tax and of spending measures. In two memo lines the OBR also splits the direct effect between decisions on and off HM Treasury's policy decisions table. - The policy workbook's Table 1.1 (36 rows): the total, direct and indirect effects of Government decisions, the direct effects of scorecard and of non-scorecard decisions, and the total of expenditure measures. - Chapter 3's text (3 rows): paragraph 3.3's spending policies, paragraph 3.4's tax policy changes and paragraph 3.5's fall in departmental underspends, a part of the indirect effects. -The OBR sums its receipts heads, from air passenger duty to visa fees, into the workbook's total of receipts measures (6 rows, cells I490:N490, rows 2190–2195 counting from 0), so that total keeps `revenue_change`, the receipts line `models.py` keeps apart from PSNB (delta review of 8b7b23db, finding 1). Table 3.1 gives the same figures, in £ billion with the opposite sign, as the direct effect of tax measures, and those rows take `decisions_effect_on_borrowing`. Table 3.1's memo on the current budget deficit (6 rows) now proposes `fiscal_balance_change`, and its sign convention adds that, on this line, a positive value increases that deficit. The workbook's 288 lines by tax and spending head, Table 3.1's groups of measures, Table 3.3's costings and the prose costings of measures keep `revenue_change`. Paragraphs 3.20, 3.21 and 3.43 give four policies' effects on borrowing, and their seven rows already took `decisions_effect_on_borrowing`. No value or printed sign changed. Counting from 0 in the 3 November order, the moved rows are 2352–2357, 2490–2519, 2521, 2527, 2533–2563 and 2630–2641, and the memo rows 2642–2647. The contract checker reports 2,797 rows, 0 problems, and a re-run of `stage.py` reproduces the gzip byte for byte. +The OBR sums its receipts heads, from air passenger duty to visa fees, into the workbook's total of receipts measures (6 rows, cells I490:N490, rows 2190–2195 counting from 0). We keep that total on `revenue_change`, the receipts line that `models.py` keeps apart from PSNB (delta review of 8b7b23db, finding 1). In Table 3.1 the OBR gives the same figures, in £ billion with the opposite sign, as the direct effect of tax measures, and we move those rows to `decisions_effect_on_borrowing`. + +We now propose `fiscal_balance_change` for Table 3.1's memo on the current budget deficit (6 rows), and add to its sign convention that a positive value on this line increases that deficit. We keep `revenue_change` for the workbook's 288 lines by tax and spending head, Table 3.1's groups of measures, Table 3.3's costings and the prose costings of measures. In paragraphs 3.20, 3.21 and 3.43 the OBR gives four policies' effects on borrowing, and we had already put their seven rows on `decisions_effect_on_borrowing`. We changed no value or printed sign. + +Counting from 0 in the 3 November order, we moved rows 2352–2357, 2490–2519, 2521, 2527, 2533–2563 and 2630–2641 (81 rows) and gave the memo rows 2642–2647 the proposed `fiscal_balance_change`. The contract checker reports 2,797 rows, 0 problems, and when we re-run `stage.py` it reproduces the gzip byte for byte. `scorecard_db/signs.py` now reads the workbook's legend on both metrics: on `revenue_change` it runs the way PolicyEngine's sign does, and on `decisions_effect_on_borrowing` the other way. It reads the six workbook total rows whose sign convention starts "mixed:", and paragraph 3.6's two absolute averages, as having no single sign. -Review B also noted that rows 2520, 2648 and 2649 (paragraphs 3.2 and 3.6) carry `decisions_effect_on_borrowing` as well, so the family has 91 rows on that metric: the 81 rule 10 moved, the seven of paragraphs 3.20, 3.21 and 3.43, and these three. +Review B also noted that we had put rows 2520, 2648 and 2649 (paragraphs 3.2 and 3.6) on `decisions_effect_on_borrowing`. So we use that metric on 91 rows in the family: the 81 we moved under rule 10, the seven of paragraphs 3.20, 3.21 and 3.43, and these three. -**Text that restates a Table 3.1 group (4 November, review B finding 12, option (a)).** Text restating a Table 3.1 group or measure as an effect on borrowing keeps that row's metric, and Table 3.1's "A positive sign implies an increase in borrowing" is a sign convention, not a statement that a group is the decisions' effect on PSNB. So paragraph 3.11's "increase borrowing by £5 billion in 2026-27" (Table 3.1's "Other departmental spending changes", rounded) keeps `revenue_change`, as that group's rows do. +**Text that restates a Table 3.1 group (4 November, review B finding 12, option (a)).** Where the OBR restates a Table 3.1 group or measure in its text as an effect on borrowing, we give it that row's metric. In Table 3.1 the OBR sets a sign convention ("A positive sign implies an increase in borrowing"), but it doesn't call any group the decisions' effect on PSNB. So we keep `revenue_change` for paragraph 3.11's "increase borrowing by £5 billion in 2026-27" (Table 3.1's "Other departmental spending changes", rounded), as for that group's rows. ## Reviews A and B, 4 November 2026 -Review B (issue comments 6635496734 to 6635499259 on #231) approved this family with fixes, and review A's key harmonisation (6635870934) touched its rows. `review_fixes_nov4()` in `stage.py` applies them after `rule10()`, with indices in the 4f424b9 order; no printed value changed. Rows: 2,797 + 1 restored + 2 added = 2,800. - -1. **The corrected PDF (finding 1):** a fourth manifest line; row 2673's `reform_hint` and note give the corrected date; the rows stay on the 28 October file. -2. **Parts and wholes on the total-decisions key (finding 2, review A's K-1 and H-1):** the totals are `measure_total` with no `component`; the workbook's direct and indirect effects, its total of receipts measures and its 288 lines by head, Table 3.1's direct and indirect effects, and paragraphs 3.2 and 3.5 are `measure_component`, each labelled as its primary prints it. The workbook's scorecard and non-scorecard rows are `measure_total` on their own keys. Table 3.1's total, indirect and current budget memo rows take `source_model: obr_efo_forecast`. -3. **Receipts-only figures (finding 3, review A's K-5 and K-7(g)):** paragraph 3.17's £0.4bn and paragraph 3.19's electricity VAT figures are receipts components of their lines; the tax administration and temporary business rates groupings take `_receipts` keys; Table 3.3's online marketplace rows are line 46's whole costing. -4. **Chapters 1, 2 and 4–6 and paragraph 7.20 (finding 4):** not staged, each listed with its reason (the coordinator's call). -5. **Paragraph 3.20's financing requirement (finding 5):** restored on `cash_requirement_change`, under the rule for lending and financing above. -6. **Splits on the split keys (finding 6):** the workbook's total of expenditure measures and paragraph 3.4's tax line are whole rows on the spending- and tax-decisions keys. -7. **Lines 70 and 71 (finding 7):** the carbon price package's scope; footnote 9's £0.8bn keyed to line 70 as its current AME. -8. **More parts (finding 8):** paragraph 3.20's gross debt and additional income, paragraph 3.12's two pension heads and paragraph 3.19's AME to local authorities (now `revenue_change`, as worded) are components. Table 3.3's four rating rows that name a part are components too (H-1). -9. **Two spending scopes (finding 9)** are restated. -10. **Box 3.1 (finding 10):** the whole sentence, from the coordinator's reading of the 28 October PDF. -11. **Box 3.1's counts (finding 11):** staged. -12. **Paragraph 3.11 (finding 12):** option (a), above. -13. **Averaging intervals (finding 13):** row 2529 averages from 2027-28; row 2522's interval is not stated. -14. **Civil service programme (finding 14):** membership probable, not established. -15. **The OBR's residual groups (finding 15)** keep `obr` in their keys. -16. **Stale and run-together text (finding 16):** notes and these NOTES corrected; `review_fixes_nov4()` inserts the missing spaces in every field but quotes, values and locators. - -Review A's K-2 makes row 2522 the defence investment plan's RDEL part, with the sign convention "positive = cost to the Exchequer (higher RDEL spending)"; K-6 keeps RF's RPI-uprating option apart from this family's fuel duty freezes key; K-7(d) gives paragraph 3.51's two simulations conventions `signs.py` reads; and H-6(b), the coordinator's call, makes paragraph 3.28's revision a context row with a null key and no `recosting_of`. +Review B (issue comments 6635496734 to 6635499259 on #231) approved this family with fixes, and review A's key harmonisation (6635870934) touched its rows. `review_fixes_nov4()` in `stage.py` applies them after `rule10()`, with indices in the 4f424b9 order, and we changed no printed value. Rows: 2,797 + 1 restored + 2 added = 2,800. + +1. **The corrected PDF (finding 1):** we added a fourth manifest line and give the corrected date in row 2675's `reform_hint` and note; we keep the rows on the 28 October file. +2. **Parts and wholes on the total-decisions key (finding 2, review A's K-1 and H-1):** + - We mark the totals `measure_total`, with no `component`. + - We mark the workbook's direct and indirect effects, its total of receipts measures and its 288 lines by head, Table 3.1's direct and indirect effects, and paragraphs 3.2 and 3.5 as `measure_component`. We set each part's `component` from the OBR's wording, adding the fiscal head for workbook lines and the paragraph number for paragraphs 3.2 and 3.5. + - We mark the workbook's scorecard and non-scorecard rows `measure_total` on their own keys, and give Table 3.1's total, indirect and current budget memo rows `source_model: obr_efo_forecast`. +3. **Receipts-only figures (finding 3, review A's K-5 and K-7(g)):** we mark paragraph 3.17's £0.4bn and paragraph 3.19's electricity VAT figures as receipts components of their lines. We gave the tax administration and temporary business rates groupings `_receipts` keys, and stage Table 3.3's online marketplace rows as line 46's whole costing. +4. **Chapters 1, 2 and 4–6 and paragraph 7.20 (finding 4):** we didn't stage them, and list each with its reason (the coordinator's call). +5. **Paragraph 3.20's financing requirement (finding 5):** we restored it on `cash_requirement_change`, under the rule for lending and financing above. +6. **Splits on the split keys (finding 6):** we stage the workbook's total of expenditure measures and paragraph 3.4's tax line as whole rows on the spending- and tax-decisions keys. +7. **Lines 70 and 71 (finding 7):** we give the carbon price package's scope, and key footnote 9's £0.8bn to line 70 as its current AME. +8. **More parts (finding 8):** we mark as components paragraph 3.20's gross debt and additional income, paragraph 3.12's two pension heads and paragraph 3.19's AME to local authorities (now `revenue_change`, as worded). We also mark as components Table 3.3's four rating rows in which the OBR names a part (H-1). +9. **Two spending scopes (finding 9):** we restated them. +10. **Box 3.1 (finding 10):** we stage the whole sentence, from the coordinator's reading of the 28 October PDF. +11. **Box 3.1's counts (finding 11):** we staged them. +12. **Paragraph 3.11 (finding 12):** we took option (a), above. +13. **Averaging intervals (finding 13):** row 2529 averages from 2027-28; the OBR doesn't state row 2522's interval. +14. **Civil service programme (finding 14):** we call its membership probable, not established. +15. **The OBR's residual groups (finding 15):** they keep `obr` in their keys. +16. **Stale and run-together text (finding 16):** we corrected the row notes and these NOTES, and `review_fixes_nov4()` inserts the missing spaces in every field but quotes, values and locators. + +Under review A's K-2 we make row 2522 the defence investment plan's RDEL part, with the sign convention "positive = cost to the Exchequer (higher RDEL spending)". Under K-6 we keep RF's RPI-uprating option apart from this family's fuel duty freezes key. Under K-7(d) we give paragraph 3.51's two simulations conventions that `signs.py` reads. Under H-6(b), the coordinator's call, we make paragraph 3.28's revision a context row with a null key and no `recosting_of`. diff --git a/sources/harvest-uk-ab2026-2026-10-29/uk_rf_ab2026/NOTES.md b/sources/harvest-uk-ab2026-2026-10-29/uk_rf_ab2026/NOTES.md index dd0604653..78ffd7158 100644 --- a/sources/harvest-uk-ab2026-2026-10-29/uk_rf_ab2026/NOTES.md +++ b/sources/harvest-uk-ab2026-2026-10-29/uk_rf_ab2026/NOTES.md @@ -1,24 +1,24 @@ # uk_rf_ab2026: Resolution Foundation, Autumn Budget 2026 -**Working files.** `stage.py`, the file meant wherever these notes say `stage.py`, lives on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/rf/uk_rf_ab2026/stage.py`; this branch carries only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz`. The lane's local primaries folder (`.ab2026-primaries/uk_rf_ab2026/`, which the `--check-primaries` command below reads) was never committed: the repository carries no primary bytes, so fetch each primary from its `manifest.jsonl` URL and check its sha256. +We keep `stage.py` on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/rf/uk_rf_ab2026/stage.py`. We commit only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz` to this branch, and no primary bytes to the repository. The lane never committed its local primaries folder (`.ab2026-primaries/uk_rf_ab2026/`, which the `--check-primaries` command below reads), so fetch each primary from its `manifest.jsonl` URL and check its sha256. -Staged 2026-10-29 by `stage.py`, which holds every row as data and rewrites `claims_staged.jsonl.gz` byte for byte. Source slug `resolution_foundation`. +We staged this family on 2026-10-29 with `stage.py`. We wrote every row into it as data, and it rewrites `claims_staged.jsonl.gz` byte for byte. We use source slug `resolution_foundation`. -- **Rows:** 766 (767 before the review dropped row 397; see "Review fixes" at the end). +- Rows: 766 (767 before we dropped row 397 after review; see "Review fixes" at the end). - Text: 212 rows from the two press releases, the Spread thin page, both versions of the Spread thin PDF and one comment. - Workbook: 554 rows from the chart-data workbook RF links from the Spread thin page. -- **Primaries:** 8 (manifest.jsonl): the seven the lane fetched, plus the Spread thin page's 4 November version, which no row cites (README rule 7; see "The replaced PDF"). -- **source_model:** +- Primaries: 8 (manifest.jsonl): the seven the lane fetched, plus the Spread thin page's 4 November version, which we cite in no row (README rule 7; see "The replaced PDF"). +- source_model: - `landman_ttm`: the IPPR tax-benefit model behind RF's distributional and living-standards figures (488 rows). - `arithmetic`: RF's sums, averages and energy-bill construction on OBR, HMT and Ofgem inputs (148). - `survey_microdata`: LCF, NEED and HBAI analysis, including HBAI outturn years (42). - - Restated rows carry the original engine: `obr_efo_forecast` (42), `hmt_costing_obr_certified` (38), `hm_government_publication` (5, Government figures that aren't Table 6.1 cells, including HM Treasury's Household Energy Payment count) and `administrative_data` (1, row 38's outturn). Two own rows also carry `administrative_data` (official counts RF computes from). + - On restated rows we name the original engine: `obr_efo_forecast` (42), `hmt_costing_obr_certified` (38), `hm_government_publication` (5, Government figures that aren't Table 6.1 cells, including HM Treasury's Household Energy Payment count) and `administrative_data` (1, row 38's outturn). We also mark two own rows `administrative_data` (official counts RF computes from). `python3 stage.py --check-primaries ../../../.ab2026-primaries/uk_rf_ab2026` re-hashes every primary against the manifest. It also finds every text quote in its document (pdftotext `-layout`, `-raw` and default modes for the PDFs) and compares all 759 embedded workbook cells with the sheet XML. On 2026-10-29 it reported 0 problems. ## Primaries and access recipe -Every primary came from `curl -sSL` with the Safari User-Agent, HTTP 200, with no blocks or challenges. A second fetch at 17:02Z returned identical bytes for all seven. The HTML pages embed per-request content (rotating nonces, tokens and related-article lists), so a later fetch's sha256 can't reproduce the manifest's; for those pages the check is that every staged quote still appears on the page (review B did this on 4 November). +The lane fetched every primary with `curl -sSL` and the Safari User-Agent, and got HTTP 200 with no blocks or challenges. A second fetch at 17:02Z returned identical bytes for all seven. RF's server embeds per-request content in the HTML pages (rotating nonces, tokens and related-article lists), so a later fetch's sha256 can't reproduce the manifest's. For those pages, check instead that every staged quote still appears on the page, as review B did on 4 November. | primary | sha256 | bytes | fetched (UTC) | rows | |---|---|--:|---|--:| @@ -31,53 +31,53 @@ Every primary came from `curl -sSL` with the Safari User-Agent, HTTP 200, with n | Comment, *It's good to be flexible* (Nye Cominetti, 29 Oct) | `3510a245…62fcfc` | 131,349 | 16:20:03 | 1 | | Spread thin publication page, 4 November version (review B's fetch) | `d81969ef…0b0259` | 154,801 | 4 Nov | 0 | -- **HTML pages:** the stager quotes each page's own article text. Locators count body paragraphs after the dateline (press releases and comment) or name the bullet list ("measures bullet 2", "Key findings bullet 3"). -- **PDFs:** read with `pdftotext -layout`, and with `-raw` for the two-column boxes. Locators give the PDF page, which matches the printed folio, and the paragraph, bullet, box, footnote or figure label. -- **Workbook:** one sheet per figure (F1 to F23, plus "Extra spending stats"). - - `value_raw` is the cell's stored `` string from the sheet XML, at full precision, not the rounded display: F8's `_-"£"* #,##0_-` shows 145.33 as £145. - - The quote is "series label | category | stored value". - - Locators are `sheet F8, cell J8`. - - Each figure's plotted ranges come from the chart objects in the workbook. Figure 21's chart object is `chart34.xml` (grouped, plotting AD66:AD74); its column AD also carries RF's own marker, "**USED IN CHART" (AD60). Figure 8's column J (the net change) is staged though the chart doesn't plot it. - - Chart rows name their series in `conditions.series`. Only the rule 9 parts (Figure 8's two VAT-cut columns and p.17's extension) also carry `conditions.component`. -- **Local copy:** the assignment's copy of `Spread-thin-1.pdf` (downloaded 08:46Z) hashes to the same `a669f39d…`, so v1 rows cite the fresh fetch. +- **HTML pages.** The stager quotes each page's own article text. In our locators we count body paragraphs after the dateline (press releases and comment) or name the bullet list ("measures bullet 2", "Key findings bullet 3"). +- **PDFs.** We read them with `pdftotext -layout`, and with `-raw` for the two-column boxes. In our locators we give the PDF page, which matches the printed folio, and the paragraph, bullet, box, footnote or figure label. +- **Workbook.** RF gives one sheet per figure (F1 to F23, plus "Extra spending stats"). + - For `value_raw` we take the cell's stored `` string from the sheet XML, at full precision, not the rounded display: with F8's `_-"£"* #,##0_-` RF shows 145.33 as £145. + - We write the quote as "series label | category | stored value". + - We write locators as `sheet F8, cell J8`. + - We take each figure's plotted ranges from the chart objects in the workbook. RF draws Figure 21 from chart object `chart34.xml` (grouped, plotting AD66:AD74), and marks column AD with its own "**USED IN CHART" (AD60). We stage Figure 8's column J (the net change), though RF's chart doesn't plot it. + - We name each chart row's series in `conditions.series`. We add `conditions.component` only on the rule 9 parts (Figure 8's two VAT-cut columns and p.17's extension). +- **Local copy.** The assignment's copy of `Spread-thin-1.pdf` (downloaded 08:46Z) hashes to the same `a669f39d…`, so in v1 rows we cite the fresh fetch. ## The replaced PDF (README rule 7) RF replaced the briefing on the morning of 29 October. -- **v1, `Spread-thin-1.pdf`:** PDF CreationDate 04:33 EDT; server Last-Modified 08:37:45 GMT. The URL still serves it, but the landing page no longer links it. -- **v2, `Spread-thin.pdf`:** CreationDate 08:13 EDT; Last-Modified 12:14:43 GMT. The landing page links this version. -- **Text changes, from a page-by-page word diff:** - - p.5: "the poorest half gain on average £158 … a paltry £14 more than the £144" became £145 / £15 / £130. - - p.19: "(equivalent to an average £158 for poorer households, compared to £144 for richer households)" became "(equivalent to under £145 on average for poorer households, compared to over £130 for richer households – a gap of less than £15)". - - Figure 8: the Council Tax rebate bars fall from about £100 to about £88 per household (vigintile 2); the other components are unchanged; the axis top moved accordingly, from +£180 / +3.6% to +£160 / +3.2%. - - Figure 4: the x-axis gains 2031. - - p.30: "restated the last Keir Starmer’s commitment" lost "the last". +- **v1, `Spread-thin-1.pdf`:** PDF CreationDate 04:33 EDT; server Last-Modified 08:37:45 GMT. RF still serves it at the URL, but its landing page no longer links it. +- **v2, `Spread-thin.pdf`:** CreationDate 08:13 EDT; Last-Modified 12:14:43 GMT. RF's landing page links this version. +- **Text changes, from our page-by-page word diff:** + - p.5: RF changed "the poorest half gain on average £158 … a paltry £14 more than the £144" to £145 / £15 / £130. + - p.19: RF changed "(equivalent to an average £158 for poorer households, compared to £144 for richer households)" to "(equivalent to under £145 on average for poorer households, compared to over £130 for richer households – a gap of less than £15)". + - Figure 8: RF lowered the Council Tax rebate bars from about £100 to about £88 per household (vigintile 2) and left the other components unchanged. It moved the axis top accordingly, from +£180 / +3.6% to +£160 / +3.2%. + - Figure 4: RF added 2031 to the x-axis. + - p.30: RF dropped "the last" from "restated the last Keir Starmer’s commitment". - Elsewhere the lines only reflow. -- **Which version rows cite:** v2 is the source for every PDF row except the three p.5 v1 rows (£158, £14, £144), which carry `conditions.document_version: "v1 (superseded)"`. -- **Workbook:** uploaded at 12:16:43 GMT, it matches v2. Its Figure 8 net gains average about £145 over vigintiles 2–10. -- **Landing page:** its Key findings still printed v1's "gains £158 on average, against £144 for the richer half" at both 29 October fetches. Those two rows (56–57) cite the 29 October bytes and carry `document_version: "29 October page text (superseded by 4 November)"`. -- **The landing page changed by 4 November.** Review B's 4 November fetch (sha256 `d81969ef…0b0259`, 154,801 bytes, JSON-LD `dateModified` 2026-11-04T09:21:48Z) prints v2's £145 / £130 in Key findings bullet 3. It is manifest line 8 (README rule 7), with no rows: v2 rows 74 and 76 already carry £145 / £130, so they aren't staged again. "By", not "on": `dateModified` bounds the edit only from above. +- **Which version we cite.** We cite v2 for every PDF row except the three p.5 v1 rows (£158, £14, £144), on which we set `conditions.document_version: "v1 (superseded)"`. +- **Workbook.** RF uploaded it at 12:16:43 GMT, and it matches v2. Its Figure 8 net gains average about £145 over vigintiles 2–10. +- **Landing page.** In Key findings RF still printed v1's "gains £158 on average, against £144 for the richer half" at both 29 October fetches. Those two rows (56–57) cite the 29 October bytes, and we set `document_version: "29 October page text (superseded by 4 November)"` on them. +- **RF changed the landing page by 4 November.** In review B's 4 November fetch (sha256 `d81969ef…0b0259`, 154,801 bytes, JSON-LD `dateModified` 2026-11-04T09:21:48Z) RF prints v2's £145 / £130 in Key findings bullet 3. We record it as manifest line 8 (README rule 7), with no rows: we already stage £145 / £130 in v2 rows 74 and 76. From RF's `dateModified` (09:21Z on 4 November) we can tell only that RF had made the edit by then. ## Harvest rules 8 and 9 -The coordinator added two rules to the AB2026 README during this family's staging, and `pipeline/ab2026_harvest_contract.py` now checks rule 9's rows. This family follows both: +The coordinator added two rules to the AB2026 README while we staged this family, and `pipeline/ab2026_harvest_contract.py` now checks rule 9's rows. We follow both: -- **Rule 8: stage estimates, not design.** A row is a number someone estimated, forecast or counted about the effect of a measure or of the Budget. - - A measure's own parameters are design: rates, thresholds, payment amounts, dates and durations. So are spending allocations to programmes, financing remits and contingent liabilities. - - Design goes in `reform_hint` or `conditions`, never as rows. - - A worked example's inputs are design too; its computed outcome is a row. -- **Rule 9: a component of one measure keeps that measure's key.** The row carries `conditions.component` and `conditions.row_type: measure_component`. Proposed package keys are only for scopes that cross Table 6.1 lines, or have none. +- **Rule 8: stage estimates, not design.** We stage as a row only a number someone estimated, forecast or counted about the effect of a measure or of the Budget. + - Under rule 8 we treat a measure's own parameters as design: rates, thresholds, payment amounts, dates and durations. We treat spending allocations to programmes, financing remits and contingent liabilities the same way. + - We put design in `reform_hint` or `conditions`, never in rows. + - We treat a worked example's inputs as design too, and stage its computed outcome as a row. +- **Rule 9: we key a component of one measure to that measure.** We give the row `conditions.component` and `conditions.row_type: measure_component`. We proposed keys only for scopes that cross Table 6.1 lines or have no line. -"Not staged" below lists the 36 design rows rule 8 removed and where their facts went. The electricity VAT cut's "original" six months and three-month "extension" are parts of Table 6.1 line 2, so their 39 rows carry line 2's key as measure components. +Under "Not staged" below we list the 36 design rows we removed under rule 8 and where their facts went. RF splits the electricity VAT cut into an "original" six months and a three-month "extension", both parts of Table 6.1 line 2, so we key their 39 rows to line 2 as measure components. -## What the family stages +## What we stage -- **Text:** every number RF prints that estimates, forecasts or counts the effect of the Budget's measures or of the Budget as a whole. That covers costs and yields, households affected, gains and losses, bill changes, distributional and poverty effects, and the public finances after the Budget. - - The text rows also include the living-standards and poverty projections and the energy-bill context the cost-of-living measures address, as `measure_key: null` context. - - Within one document, a number printed twice with the same meaning gets one row, at the occurrence that states its period or scope most completely. For example, p.16 states the £2,192 cap level as Q1 2027, which p.5 calls "next year", so the row cites p.16. - - Different documents' statements of the same figure are each staged (the AB2025 precedent; dedup belongs to the ingest). -- **Workbook:** the chart data behind figures of RF's own estimates. +- **Text.** We stage every number RF prints that estimates, forecasts or counts the effect of the Budget's measures or of the Budget as a whole. That covers costs and yields, households affected, gains and losses, bill changes, distributional and poverty effects, and the public finances after the Budget. + - We also stage, as `measure_key: null` context, the living-standards and poverty projections and the energy-bill context that the cost-of-living measures address. + - Where RF prints a number twice with the same meaning in one document, we stage one row, at the occurrence where RF states its period or scope most completely. For example, RF dates the £2,192 cap level Q1 2027 on p.16 and calls it "next year" on p.5, so we cite p.16 in the row. + - Where a producer states the same figure in different documents, we stage each (the AB2025 precedent). At ingest we merge rows 26, 27, 60, 132 and 201 into the claim of another printing of the same RF figure. +- **Workbook.** We stage the chart data behind the figures in which RF gives its own estimates. - F7 (energy bill and support by year). - F8 (temporary AB26 policies by vigintile). - F9 (LHA unfreeze and Your First Home by decile). @@ -86,11 +86,11 @@ The coordinator added two rules to the AB2026 README during this family's stagin - F21 (median non-pensioner income growth since 2023-24). - F22 (non-pensioner income growth by vigintile and year). - F23 (change in AHC poverty since 2023-24). -- **Structural zeros, not staged:** +- **Structural zeros, which we didn't stage:** - F7 support components in years when the support didn't exist (for example the Energy Price Guarantee in 2019-20). - The 2023-24 zero base of F21 and F23. - F22's vigintile 1, a zero placeholder for the excluded bottom 5 per cent. - - F8 and F20 leave vigintile 1 blank. + - In F8 and F20 RF leaves vigintile 1 blank. ## Coverage tally @@ -100,7 +100,8 @@ The coordinator added two rules to the AB2026 README during this family's stagin | measure_key | rows | |---|--:| -| null (proposed key, or context) | 532 | +| null (context) | 332 | +| the 19 keys we proposed (below) | 200 | | `ab2026__package_household_energy_vat_and_fuel_duty` | 54 | | `ab2026__electricity_vat_zero_rate_to_july_2027` | 42 (39 of them measure components) | | `ab2026__household_energy_payment_100_bands_a_to_d` | 31 | @@ -116,46 +117,46 @@ The coordinator added two rules to the AB2026 README during this family's stagin | `ab2026__founders_relief_cgt` | 1 | | `ab2026__defence_investment_plan_further_savings` | 1 | -The 532 null rows split into 200 with a `proposed_measure_key` (listed below) and 332 context rows. The context rows cover fiscal aggregates, energy bill levels, living standards and poverty. They also include the three pre-AB26 component columns of F20 (57 rows: "pre-AB26 Direct tax changes", "pre-AB26 Benefit changes" and "pre-AB26 Indirect tax changes"), which score earlier events' measures. +We key 200 rows to the 19 keys below, which we proposed and registered in tranche 2, and stage 332 as context. As context we stage fiscal aggregates, energy bill levels, living standards and poverty. We also stage as context F20's three pre-AB26 component columns (57 rows: "pre-AB26 Direct tax changes", "pre-AB26 Benefit changes" and "pre-AB26 Indirect tax changes"), in which RF scores earlier events' measures. ## Conventions -- **Sign.** Each change row's `conditions.sign_convention` states it, and no value is re-signed from what the primary prints or stores. Figures 21 and 22 use "positive = growth"; the text growth rows and row 117 use "positive = gain to households"; row 10 uses "positive = saving (a reduction in the household's bills)". - - Household rows use "positive = gain to households". Text printed as a fall ("£2,000 lower", "fall by 3.0 per cent", "worse off by an average of £61") carries a negative value, as in AB2025. - - Exchequer rows say "positive = cost" or "positive = yield", matching how RF writes the figure. - - Figure 10 rows keep RF's plotted sign, positive = raises revenue. RF flipped the OBR's borrowing sign (workbook F10 I15, "Signs flipped from source"). - - F7 support components are negative, meaning a reduction in the typical bill, as stored. -- **Units.** £ values are in pounds (`gbp`), and `normalization` records the scaling, for example "£ billion × 1e9 → £". - - Per cent figures are `percent`; shares and "one-in-five"-style fractions are `share`. - - The stager multiplies workbook cells that store fractions (F8 K, F20 M, F21, F22) by 100 to give `percent`. F23 already stores percentage points. +- **Sign.** We state the convention in each change row's `conditions.sign_convention`, and we re-signed no value from what RF prints or stores. We give Figures 21 and 22 "positive = growth", the text growth rows and row 117 "positive = gain to households", and row 10 "positive = saving (a reduction in the household's bills)". + - We give household rows "positive = gain to households". Where RF prints a fall ("£2,000 lower", "fall by 3.0 per cent", "worse off by an average of £61"), we stage a negative value, as in AB2025. + - On Exchequer rows we say "positive = cost" or "positive = yield", matching how RF writes the figure. + - On Figure 10 rows we keep RF's plotted sign, positive = raises revenue. RF flipped the OBR's borrowing sign (workbook F10 I15, "Signs flipped from source"). + - We keep F7 support components negative, as RF stores them, for a reduction in the typical bill. +- **Units.** We give £ values in pounds (`gbp`), and record the scaling in `normalization`, for example "£ billion × 1e9 → £". + - We give per cent figures as `percent`, and shares and "one-in-five"-style fractions as `share`. + - The stager multiplies workbook cells that store fractions (F8 K, F20 M, F21, F22) by 100 to give `percent`. RF already stores percentage points in F23. - **Price basis:** - F8 and the package's halves figures: cash terms. - F7, F20 and the energy text: 2026-27 prices (Spread thin fn 13). - £ living-standards figures: 2026-27 prices, as printed. - - `conditions.price_basis` records the basis on each row. + - We record the basis on each row in `conditions.price_basis`. - **Period:** - - Fiscal-year figures carry `fy` and the end year. - - The temporary package (Figure 8, the 52/48 split, the halves' averages and gap, and the £4.4bn) covers 2026-27 and 2027-28 together. Those rows carry no `fy`, `period` 2028 (the £4.4bn: 2027, as printed or read), `time_basis: fiscal_year` and `conditions.horizon`. Figure 8 shows each measure's "total cash impact" over the two years (v2 p.20; workbook F8 N35), so its rows and the text figures that restate it are `cumulative`, and the F8 £ columns' normalization reads "£ per household as stored: total cash impact over 2026-27 and 2027-28 (Figure 8 note)". - - The gas standing-charge rows (115–125) carry no `fy`: RF prints no year. They sit at period 2028 with `conditions.horizon` "year not printed: the change lands by the end of 2027 or soon after (p.18)", at medium confidence. - - Averages over a span (for example "£11 billion a year on average over the next five years") take the span's end year with `statistic: annual average`. - - Calendar figures (the 2027 average bill, decade and Parliament RHDI growth) are `annual`; Q1-2027 cap levels and "currently" figures are `point_in_time`. + - We give fiscal-year figures `fy` and the end year. + - RF's temporary package (Figure 8, the 52/48 split, the halves' averages and gap, and the £4.4bn) covers 2026-27 and 2027-28 together. We give those rows no `fy`, `period` 2028 (the £4.4bn rows: 2027, which RF prints on p.19 as "2025-26 and 2026-27" for row 127 and which we read from p.19 for rows 34 and 54), `time_basis: fiscal_year` and `conditions.horizon`. In Figure 8 RF shows each measure's "total cash impact" over the two years (v2 p.20; workbook F8 N35). So we mark its rows, and the text figures that restate it, `cumulative`. For the F8 £ columns, we record "£ per household as stored: total cash impact over 2026-27 and 2027-28 (Figure 8 note)" in `normalization`. + - RF prints no year for the gas standing-charge rows (115–125), so we give them no `fy`. We put them at period 2028 with `conditions.horizon` "year not printed: the change lands by the end of 2027 or soon after (p.18)", at medium confidence. + - For averages over a span (for example "£11 billion a year on average over the next five years"), we use the span's end year with `statistic: annual average`. + - We mark calendar figures (the 2027 average bill, decade and Parliament RHDI growth) `annual`, and Q1-2027 cap levels and "currently" figures `point_in_time`. - **Value kinds.** RF prints point estimates throughout. - - "under", "up to" and "less than" give `range_high`; "over" and "just over" give `range_low`. - - Totals over several years and growth since a base year are `cumulative`: 195 rows, including Figure 8's 114 and the 11 text rows that restate Figure 8 or the £4.4bn. -- **Income groups.** `income_group` uses the AB2025 spellings `vigintile_N`, `decile_N`, `quintile_N`, `bottom_half` and `top_half`, the last two for RF's "poorer / richer half". The AB2025 NOTES flagged `bottom_half` / `top_half` for an ingest ruling. + - We read "under", "up to" and "less than" as `range_high`, and "over" and "just over" as `range_low`. + - We mark totals over several years and growth since a base year `cumulative`: 195 rows, including Figure 8's 114 and the 11 text rows that restate Figure 8 or the £4.4bn. +- **Income groups.** In `income_group` we use the AB2025 spellings `vigintile_N`, `decile_N`, `quintile_N`, `bottom_half` and `top_half`, the last two for RF's "poorer / richer half". In the AB2025 NOTES we flagged `bottom_half` / `top_half` for an ingest ruling. - **Geography.** - - F8 rows are "UK", with `modelling_geography` quoting RF's note (Council Tax and VAT cuts modelled for GB, Fuel Duty for the UK). - - F9 and the Your First Home and LHA rows are England (Figure 9). - - Energy bills are GB. + - We set F8 rows to "UK", with `modelling_geography` quoting RF's note (Council Tax and VAT cuts modelled for GB, Fuel Duty for the UK). + - We set F9 and the Your First Home and LHA rows to England (Figure 9). + - We set energy bills to GB. - **Baselines.** - - F20's net and pre-AB26 columns, and the Parliament-wide text figures (99 rows), use `rf_start_of_parliament_policy` ("relative to the baseline inherited from Chancellor Hunt"). - - Every AB26 measure row is null: pre-Budget policy, current law at the scoring date. That includes F20's AB26 component columns, increments within RF's stack. + - For F20's net and pre-AB26 columns, and the Parliament-wide text figures (99 rows), we use `rf_start_of_parliament_policy` ("relative to the baseline inherited from Chancellor Hunt"). + - We leave the baseline null on every AB26 measure row: pre-Budget policy, current law at the scoring date. That includes F20's AB26 component columns, increments within RF's stack. - **Attribution.** - - `restated` (86 rows): HM Treasury Table 6.1 costings, OBR forecasts, the Government's count of carers and HM Treasury's count of Household Energy Payment households (rows 55 and 72, which RF repeats uncited). `note` names the original on each. + - `restated` (86 rows): HM Treasury Table 6.1 costings, OBR forecasts, the Government's count of carers and HM Treasury's count of Household Energy Payment households (rows 55 and 72, which RF repeats uncited). We name the original in `note` on each. - `own` (680): everything RF computes, including its sums and averages of official figures. - - `benchmark_class` is `administrative_fact` for outturns and official counts (53 rows), including the 2024-25 HBAI outturn year in F21–F23, and `different_model` otherwise. -- **Worked examples (mode 3):** the computed outcomes only (rule 8). - - The p.25 Pension Tax Allowance comparison: £34 Income Tax and £10 NICs for the worker, "nothing" for the pensioner. The £12,739 earnings input sits in `household_type`. + - We set `benchmark_class` to `administrative_fact` for outturns and official counts (53 rows), including the 2024-25 HBAI outturn year in F21–F23, and `different_model` otherwise. +- **Worked examples (mode 3).** We stage the computed outcomes only (rule 8). + - The p.25 Pension Tax Allowance comparison: £34 Income Tax and £10 NICs for the worker, "nothing" for the pensioner. We put the £12,739 earnings input in `household_type`. - The Box 4 Carer's Allowance cliff edge: net income of £280, then £216 a week. - The NLW worker's £1,040 a year. @@ -163,61 +164,61 @@ The 532 null rows split into 200 with a `proposed_measure_key` (listed below) an ### Registered keys -- **`ab2026__package_household_energy_vat_and_fuel_duty` (Table 6.1 lines 1, 2 and 3).** This key covers RF's "cost of living package": Figure 8's net change and per cent change, the 52 / 48 per cent split, the halves' averages and gap, and the landing page's £158 / £144. The £4.4bn costing is not on it (review A, K-4; see the proposed keys). - - RF's package is the Household Energy Payment, the electricity VAT cut ("original" six months plus the three-month "extension") and the Fuel Duty delay. - - Table 6.1 line 2's title covers the whole cut ("from 1 October 2026 to 1 July 2027"), and its costs (£755m in 2026-27, £305m in 2027-28) include the original six months. - - The registry's own modifier note gives the same reading: "about 57% of 2026-27 … about 22% of 2027-28". So RF's package equals lines 1, 2 and 3 exactly. - - The £4.4bn rows (34, 54, 127) count Fuel Duty only at the share accruing directly to households (fn 19, `costing_basis`), so their scope is not exactly lines 1–3 (rule 3) and they carry the proposed key `ab2026__package_household_energy_vat_and_household_share_of_fuel_duty`. +- **`ab2026__package_household_energy_vat_and_fuel_duty` (Table 6.1 lines 1, 2 and 3).** We key RF's "cost of living package" here. That covers Figure 8's net change and per cent change, the 52 / 48 per cent split, the halves' averages and gap, and the landing page's £158 / £144. We moved the £4.4bn costing off it (review A, K-4; see the keys we proposed). + - RF's package comprises the Household Energy Payment, the electricity VAT cut ("original" six months plus the three-month "extension") and the Fuel Duty delay. + - HM Treasury's title for Table 6.1 line 2 covers the whole cut ("from 1 October 2026 to 1 July 2027"), and its costs (£755m in 2026-27, £305m in 2027-28) include the original six months. + - In the registry's own modifier note we read the line the same way: "about 57% of 2026-27 … about 22% of 2027-28". So RF's package equals lines 1, 2 and 3 exactly. + - In the £4.4bn rows (34, 54, 127) RF counts Fuel Duty only at the share accruing directly to households (fn 19, `costing_basis`). Their scope therefore differs from lines 1–3 (rule 3), and we key them to `ab2026__package_household_energy_vat_and_household_share_of_fuel_duty`. - **`ab2026__electricity_vat_zero_rate_to_july_2027` (line 2):** - **The whole cut:** the £40 "total nine-month benefit of the cut", and Figure 7's "Electricity VAT cut" series split by financial year (£26.77 in 2026-27, £12.80 in 2027-28). - - **Measure components (rule 9, `row_type: measure_component`):** Figure 8's "Energy VAT cut (original)" (the six months to the end of Q1 2027, 19 vigintile rows), "Energy VAT cut extension" (April to 1 July 2027, 19 rows), and p.17's extension "worth £12". `conditions.component` and `measure_part` name the part. Never sum them with a whole-cut row. -- **`ab2026__household_energy_payment_100_bands_a_to_d`:** Figure 8's "Council Tax rebate", the eligibility shares and counts, and the £2.6bn cost. The £100 payment itself is design (rule 8). + - **Measure components (rule 9, `row_type: measure_component`):** Figure 8's "Energy VAT cut (original)" (the six months to the end of Q1 2027, 19 vigintile rows), "Energy VAT cut extension" (April to 1 July 2027, 19 rows), and p.17's extension "worth £12". We name the part in `conditions.component` and `measure_part`. Never sum them with a whole-cut row. +- **`ab2026__household_energy_payment_100_bands_a_to_d`:** Figure 8's "Council Tax rebate", the eligibility shares and counts, and the £2.6bn cost. We treat the £100 payment itself as design (rule 8). - **`ab2026__fuel_duty_5p_cut_extended_and_rpi_rise_deferred`:** Figure 8's "Fuel Duty freeze extension", Figure 10's "Fuel Duty" series (equal to line 3), the "just under £1.4 billion" cost and the "around half" household incidence. -- **`ab2026__pension_tax_allowance`:** F20's "Pensioner Personal Allowance" (the workbook's label; the PDF says "Pension Tax Allowance"), the £150m cost and the pensioner who "will pay nothing". -- **`ab2026__machine_games_duty_higher_stake_25pc`:** F20's "Machine Games Levy" and "around £145 million per year … through Gambling Duties … well-targeted at machine gaming" (Table 6.1 line 56 prints £145m a year from 2029-30). +- **`ab2026__pension_tax_allowance`:** F20's "Pensioner Personal Allowance" (RF's workbook label; in the PDF RF writes "Pension Tax Allowance"), the £150m cost and the pensioner who "will pay nothing". +- **`ab2026__machine_games_duty_higher_stake_25pc`:** F20's "Machine Games Levy" and "around £145 million per year … through Gambling Duties … well-targeted at machine gaming" (HM Treasury prints £145m a year on Table 6.1 line 56 from 2029-30). - **`ab2026__your_first_home_equity_loan`:** Figure 9's "Could access Your First Home", the 76 per cent top-half share, "up to 70,000 families" and the £95m. - **`ab2026__carers_allowance_earnings_taper`:** the £80m, the 48,000 carers and the Box 4 cliff edge the taper removes. -- **Restated lines.** These are each exactly one Table 6.1 line, restated by RF: +- **Restated lines.** In each of these RF restates exactly one Table 6.1 line: - Bank Surcharge (line 55): Figure 10 and "around £1.5 billion". - Writing-down allowances (line 58): Figure 10's "Capital allowances". - Founders' Relief (line 26): £300m. - Business Rates Retention (line 21): £10m and £140m. - Table 6.1 total policy decisions: £7.6bn in 2026-27. - - Defence Investment Plan line 40: RF's "addition to pay for the DIP … £6.7 billion between 2026-27 and 2031-32" equals line 40's sum (-£6,670m). The row stays at medium confidence because line 40's title reads "Commitment to further savings made at announcement on 30 June 2026". + - Defence Investment Plan line 40: RF's "addition to pay for the DIP … £6.7 billion between 2026-27 and 2031-32" equals line 40's sum (-£6,670m). We keep the row at medium confidence because HM Treasury titles line 40 "Commitment to further savings made at announcement on 30 June 2026". -### Proposed keys +### Keys we proposed -Each row below has `measure_key: null` and this key in `conditions.proposed_measure_key`. Every one crosses Table 6.1 lines or has none (rule 9). +We gave each row below `measure_key: null` and its key in `conditions.proposed_measure_key`, and in tranche 2 we registered the keys and moved them into `measure_key`. Every one crosses Table 6.1 lines or has none (rule 9). -| proposed key | rows | scope (Table 6.1 lines; RF's words) | +| key | rows | scope (Table 6.1 lines; RF's words) | |---|--:|---| | `ab2026__package_rf_parliament_permanent_tax_and_benefit_measures` | 42 | All permanent personal tax (direct and indirect) and benefit changes announced this Parliament, from the 2024 election to AB26 inclusive, against the policy inherited from Hunt. F20 "Net change in income" and "% change in income"; p.40 £320 / £2,000; p.41 0.9–2.2 per cent. Spans several events; AB26's part is lines 52, 56, 61, 62 and the gas standing charge. | -| `ab2026__gas_standing_charge_moved_to_unit_rates` | 31 | The announced abolition of the gas standing charge, "moving the £108 a year paid by the typical household onto unit rates by the end of 2027 or soon after". It costs the Exchequer nothing and has no Table 6.1 line. Covers p.18–19 (unit rates up around one-tenth; gainers and losers, by half and by children) and F20 "Remove standing charge on gas bills". The £108 sits in `reform_hint`. | +| `ab2026__gas_standing_charge_moved_to_unit_rates` | 31 | The announced abolition of the gas standing charge, "moving the £108 a year paid by the typical household onto unit rates by the end of 2027 or soon after". It costs the Exchequer nothing and has no Table 6.1 line. Covers p.18–19 (unit rates up around one-tenth; gainers and losers, by half and by children) and F20 "Remove standing charge on gas bills". We keep the £108 in `reform_hint`. | | `ab2026__package_property_and_savings_income_rates` | 25 | Table 6.1 lines 61 and 62 together: F10 "Savings & property" (their sum) and F20 "Property and savings tax rise". | | `ab2026__package_rf_ab26_permanent_personal_measures` | 19 | F20 "AB26 Changes" = lines 52, 56, 61, 62 plus the gas standing charge (the workbook's column K equals columns F + G + I + J). | | `ab2026_option__lha_unfreeze_2027_28` | 17 | Unfreezing Local Housing Allowance in 2027-28 instead of prolonging the freeze (no Table 6.1 line). Figure 9 "Would benefit if LHA were unfrozen"; "over a million families lose out, 91 per cent … in the bottom half". | | `ab2026__package_tax_decisions_since_march_2026` | 13 | The OBR's Table 3.1 "Direct effect of tax measures" since March, sign flipped. F10 "Net direct effects", the p.23 labels, and p.4 "about £5 billion annually from next year". Differs from Table 6.1's tax total (-£1,749m vs -£1,645m in 2026-27). | | `ab2026_option__tax_decisions_with_continued_fuel_duty_freeze` | 11 | The OBR's direct effect of tax measures plus keeping Fuel Duty at current rates: F10 "Net effect if Fuel Duty frozen" and its p.23 labels. On `revenue_change` (README rule 10; see "README rule 10" below). | -| `ab2026__package_tax_administration_compliance_and_debt_collection_receipts` | 7 | The OBR's Table 3.1 group "Tax administration, compliance and debt collection measures", which is the receipts of Table 6.1 lines 45 and 46 (OBR review B, finding 3; it includes timely payment for Self Assessment; RF doesn't list the lines). F10 row 52; p.4 "over £3 billion annually by 2030-31". | -| `ab2026_option__fuel_duty_continued_freeze` | 6 | Fuel Duty rates held at 52.95p a litre instead of rising to 60.59p in July 2027, relative to the announced policy: the OBR's constant-duty-rate alternative less its central forecast. F10 "… and if Fuel Duty frozen"; p.24 £4.3bn in 2029-30. A receipts cost, on `revenue_change`. For tranche 2: the OBR family holds the same alternative's £4.3bn as an effect on borrowing (paragraph 3.43, `decisions_effect_on_borrowing`), and RF states a receipts cost; the ingest settles whether they compare. | +| `ab2026__package_tax_administration_compliance_and_debt_collection_receipts` | 7 | The OBR's Table 3.1 group "Tax administration, compliance and debt collection measures", which covers the receipts of Table 6.1 lines 45 and 46 (OBR review B, finding 3; it includes timely payment for Self Assessment; RF doesn't list the lines). F10 row 52; p.4 "over £3 billion annually by 2030-31". | +| `ab2026_option__fuel_duty_continued_freeze` | 6 | Fuel Duty rates held at 52.95p a litre instead of rising to 60.59p in July 2027, relative to the announced policy: the OBR's constant-duty-rate alternative less its central forecast. F10 "… and if Fuel Duty frozen"; p.24 £4.3bn in 2029-30. A receipts cost, on `revenue_change`. At ingest we drop these rows as restatements of the OBR's forecast and keep the OBR family's £4.3bn (paragraph 3.43, row 2740) on `decisions_effect_on_borrowing`, because the OBR states its effect on borrowing. | | `ab2026__package_rf_other_tax_measures` | 6 | RF's residual in F10 "Other tax measures": OBR net direct effects less Fuel Duty, Bank Surcharge, capital allowances, savings and property, and tax administration. | -| `ab2026__package_household_energy_vat_and_household_share_of_fuel_duty` | 3 | Table 6.1 lines 1 and 2 plus the household-incidence share of line 3: RF's "£4.4 billion cost of living package", which counts Fuel Duty "at the share accruing directly to households" (fn 19; about half of line 3's £1.4bn). Lines 1 and 2 are £3,685m over 2026-27 and 2027-28. A two-year total, `cumulative` (review A, K-4). | +| `ab2026__package_household_energy_vat_and_household_share_of_fuel_duty` | 3 | Table 6.1 lines 1 and 2 plus the household-incidence share of line 3: RF's "£4.4 billion cost of living package", in which RF counts Fuel Duty "at the share accruing directly to households" (fn 19; about half of line 3's £1.4bn). Lines 1 and 2 sum to £3,685m over 2026-27 and 2027-28. A two-year total, `cumulative` (review A, K-4). | | `ab2026__package_household_energy_payment_and_electricity_vat` | 5 | Lines 1 and 2: "A £3.7 billion package of household support"; p.17 "£3.4 billion in 2026-27", "£0.3 billion in 2027-28", 45 per cent. The sums show line 2 is the whole cut; see the discrepancies below. | | `ab2026__package_government_decisions_since_march_2026` | 3 | The OBR's total effect of the Budget's decisions on borrowing, direct and indirect, including decisions not on Table 6.1: "an average of £2 billion a year" (two documents); fn 47 "no overall impact". | | `ab2026__package_rdel_decisions_excluding_household_energy_payment` | 3 | All the Budget's RDEL decisions over the Spending Review period except the Household Energy Payment: "£3 billion a year", "£9 billion top up" (two documents). A fiscal aggregate, not a programme allocation. | | `ab2026__package_rf_absolute_policy_changes` | 2 | Every policy decision counted at absolute value (RF's fiscal-event size measure): "£7 billion". RF lists no lines. | | `ab2026__package_rf_energy_bill_support_announced_at_budget` | 2 | "Support of up to £116 announced today" and "less than a fifth" of the 2027 bill rise. RF doesn't itemise the £116; it includes the £100 Household Energy Payment. | -| `ab2026_option__fuel_duty_rpi_uprating_since_2011` | 2 | Fuel Duty uprated by RPI since January 2011: "around £20 billion" a year (two documents). It mirrors the OBR family's `ab2026__package_fuel_duty_freezes_since_2010_11` (EFO paragraph 3.43): one pair of paths seen from opposite sides, kept as separate keys (review A, K-6). RF's +£20bn is the revenue the freezes forgo relative to RPI uprating, staged "positive = yield" from the option's side. | -| `ab2026__minimum_wage_uprating_april_2027` | 2 | The April 2027 minimum wage rates the Chancellor confirmed (no Table 6.1 line): "worth £1,040 a year for someone working 40 hours a week" (PDF p.12 and the comment). The rates (3.9 per cent; £13.21; 50p) sit in `reform_hint`. | -| `ab2026__gas_supplier_levy_funded_from_general_taxation` | 1 | Funding the Gas Supplier Levy from general taxation: "save the typical household £7 in 2027-28" (fn 12). No Table 6.1 line found. | +| `ab2026_option__fuel_duty_rpi_uprating_since_2011` | 2 | Fuel Duty uprated by RPI since January 2011: "around £20 billion" a year (two documents). It mirrors the OBR family's `ab2026__package_fuel_duty_freezes_since_2010_11` (EFO paragraph 3.43): one pair of paths seen from opposite sides, and we keep them on separate keys (review A, K-6). RF's +£20bn measures the revenue the freezes forgo relative to RPI uprating, which we stage "positive = yield" from the option's side. | +| `ab2026__minimum_wage_uprating_april_2027` | 2 | The April 2027 minimum wage rates the Chancellor confirmed (no Table 6.1 line): "worth £1,040 a year for someone working 40 hours a week" (PDF p.12 and the comment). We keep the rates (3.9 per cent; £13.21; 50p) in `reform_hint`. | +| `ab2026__gas_supplier_levy_funded_from_general_taxation` | 1 | Funding the Gas Supplier Levy from general taxation: "save the typical household £7 in 2027-28" (fn 12). We found no Table 6.1 line. | ## Proposed metrics -None of these is a registered Metric. Each row carries one as `proposed_metric`, reusing the AB2025 spelling where one exists: `forecast_borrowing_change`, `average_energy_bill_change`, `incidence_share`, `tax_to_gdp_ratio`, `real_spending_growth`, `spending_pressure` and `forecast_revenue_change`. The rest are new. Every unit is a registered UnitConcept. +We have registered none of these in the `Metric` enum of `scorecard_db/models.py`. We give each row one as `proposed_metric`, reusing the AB2025 spelling where one exists: `forecast_borrowing_change`, `average_energy_bill_change`, `incidence_share`, `tax_to_gdp_ratio`, `real_spending_growth`, `spending_pressure` and `forecast_revenue_change`. We coined the rest. Every unit we use appears in the registry as a UnitConcept. - **Energy:** - `typical_energy_bill`: the typical household's annual bill level. - - `average_energy_bill_change`: a change in it from prices or support combined, not from one policy. The registered `energy_bill_change` carries single-policy effects. + - `average_energy_bill_change`: a change in it from prices or support combined, not from one policy. We use the registered `energy_bill_change` for single-policy effects. - `unit_rate_change`. - `bill_rise_offset_share`. - **Public finances:** @@ -240,25 +241,25 @@ None of these is a registered Metric. Each row carries one as `proposed_metric`, - `real_income_change` (£ change in a real income statistic). - `income_change_gap`. - `net_weekly_income`. - - `gross_earnings_change` (the AB2025 spelling, which the red book family uses for the same minimum wage measure; review A, K-7(c)). + - `gross_earnings_change` (the AB2025 spelling, which we use in the red book family for the same minimum wage measure; review A, K-7(c)). ## Not staged, and why -- **Design, not estimates (rule 8; 36 rows removed after the rule arrived):** - - **Fuel Duty rates:** PR28's "7.6p rise"; p.24's 52.95p to 60.59p a litre and "over 14 per cent". The rates sit in the continued-freeze option row's `conditions.counterfactual`. - - **Minimum wage rates and counterfactual upratings:** 3.9 per cent, £13.21 and 50p (p.12 and the comment), the "on-course" 5.7 per cent and the youth rate's 10 per cent (Box 1 and the comment). The rates sit in the £1,040 rows' `reform_hint`. +- **Design, not estimates (rule 8; we removed 36 rows after the rule arrived):** + - **Fuel Duty rates:** PR28's "7.6p rise"; p.24's 52.95p to 60.59p a litre and "over 14 per cent". We keep the rates in the continued-freeze option row's `conditions.counterfactual`. + - **Minimum wage rates and counterfactual upratings:** 3.9 per cent, £13.21 and 50p (p.12 and the comment), the "on-course" 5.7 per cent and the youth rate's 10 per cent (Box 1 and the comment). We keep the rates in the £1,040 rows' `reform_hint`. - **Financing remits:** NWF £15.4bn, BBB £5bn, the combined £20.4bn (PR28 and p.6), the 0.27 per cent of GDP of investment commitments, the £5bn defence ringfence and "around a quarter" (p.33). - **Programme allocations:** - The AI package's £700m, with £165m biosecurity, £120m AI Security Institute, £140m Sovereign AI Unit, £30m local partnerships and the £15m challenge prize. - Good Growth Funds' £285m in 2027-28. - The Social and Affordable Homes Programme's £1.5bn top-up and its 10,000 homes. - - **The £108 gas standing charge:** it sits in the gas rows' `reform_hint`. - - **A worked example's input:** the £12,739 part-time earnings, which sit in the worker rows' `household_type`. + - **The £108 gas standing charge:** we keep it in the gas rows' `reform_hint`. + - **A worked example's input:** the £12,739 part-time earnings, which we keep in the worker rows' `household_type`. - **Figure 7's flat payments:** the Council Tax rebates (£100 in 2026-27, the 2022 £150) and the 2022-23 £400 Energy Bills Support Scheme, as F7 cells E7, L7 and L11. - **The 23 per cent rise in private rents since September 2023 (p.21):** an ONS statistic about no measure. -- **Charts without printed values.** The PDF's figures print no bar values except where staged: Figure 7's two 'After rebate' annotations (the two 'Net bill' labels are staged from p.17 bullet 2), Figure 10's eleven line labels and Figure 23's end labels. The exact values come from the workbook instead. Figure 5's decade labels for the 1960s–2010s are historical ONS outturn, not about the Budget; the text rows carry the 2020s values (1.4 per cent, £400). -- **Workbook sheets left out:** - - These sheets chart OBR, ONS, IMF or HMT aggregates and history, or RF's arithmetic on them. This follows AB2025, which staged these charts' headline text values instead. +- **Charts without printed values.** RF prints bar values on only three of the PDF's figures: Figure 7's two 'After rebate' annotations (we stage the two 'Net bill' labels from p.17 bullet 2), Figure 10's eleven line labels and Figure 23's end labels. We take the exact values from the workbook. In Figure 5's decade labels for the 1960s–2010s RF gives historical ONS outturn, not anything about the Budget; we stage the 2020s values (1.4 per cent, £400) in text rows. +- **Workbook sheets we left out:** + - In these sheets RF charts OBR, ONS, IMF or HMT aggregates and history, or its arithmetic on them. We follow AB2025, where we staged these charts' headline text values instead. - F1 (borrowing and debt, pre-measures). - F2 (borrowing decomposition). - F3 (gilt yields, debt interest). @@ -274,12 +275,12 @@ None of these is a registered Metric. Each row carries one as `proposed_metric`, - F18 (current-balance forecast history). - F19 (international borrowing). - "Extra spending stats". - - F12 holds marginal tax rates, which are design. - - Within the staged sheets, the family leaves out: + - In F12 RF gives marginal tax rates, which we treat as design. + - Within the staged sheets, we left out: - Working and input blocks: F9's raw winner tables, F10's HMT and OBR input rows 5–40, F21's percentile columns other than AD, and F23's level, raw and scenario blocks. - - F22's second chart (bottom 7 / middle 6 / top 7 vigintiles since 2019-20), which the briefing doesn't print. - - F23's second chart (AP72:AR73, 2029-30 and 2030-31), which the briefing doesn't print. -- **Text numbers left out:** + - F22's second chart (bottom 7 / middle 6 / top 7 vigintiles since 2019-20), which RF doesn't print in the briefing. + - F23's second chart (AP72:AR73, 2029-30 and 2030-31), which RF doesn't print in the briefing. +- **Text numbers we left out:** - **Macro forecasts and statistics:** - CPI paths (6.5 / 3.9 per cent, 43 / 22 per cent, 2.1 / 3.8 per cent, 16 / 13 per cent). - Gilt yields (44 basis points). @@ -304,7 +305,12 @@ None of these is a registered Metric. Each row carries one as `proposed_metric`, - The £72bn average of Reeves's two Budgets, which has no single period. - The 19th of 34 fiscal events (a rank). - PR28's "mid-2029" date for living standards to recover, and its ranking of the Parliament as the "second worst … since the 1950s" (a date and a rank, not an estimate of a measure). - - **Other design:** £100, bands A-D, 5p, 3p / 2p, 2 points, 3 to 6 per cent, the 31 per cent headline bank rate, 20 / 25 per cent, the £1.5bn Good Growth lending facility, the 20 per cent equity loan and 2.5 per cent deposit, the £110,000 / £140,000 income caps, £86.45, £204, 16 hours, £1 for £1, the 55 per cent UC taper, Business Rates retention shares (20 / 50 / 70 / 100 per cent), and the 3 and 3.5 per cent defence targets. + - **Other design:** + - £100, bands A-D, 5p, 3p / 2p, 2 points, 3 to 6 per cent, + - the 31 per cent headline bank rate, 20 / 25 per cent, + - the £1.5bn Good Growth lending facility, the 20 per cent equity loan and 2.5 per cent deposit, the £110,000 / £140,000 income caps, + - £86.45, £204, 16 hours, £1 for £1, the 55 per cent UC taper, + - Business Rates retention shares (20 / 50 / 70 / 100 per cent), the 3 and 3.5 per cent defence targets. - **Third-party figures:** - Anthropic's $65bn (£48bn) round and the "1.5 per cent" comparison. - The Health Foundation's £7.5bn. @@ -327,59 +333,59 @@ None of these is a registered Metric. Each row carries one as `proposed_metric`, - "around half a million low-income families" (the AB2025 two-child limit repeal). - The £24bn year-four headroom (a rounding of £23.6bn). - p.15's "£7 billion in 2029-30" (a rounding of Box 2's £6.7bn). -- **Publications checked:** - - RF's publications, press-release, comment, Top of the Charts, events and Budgets listings were all fetched at 16:19–16:20Z. - - The only other RF Budget output from 28–29 October is the launch event page, *Pleasing people and satisfying markets*. It is an embedded YouTube video with no printed figures, so the family neither hashes nor stages it. - - The comment's three slide images (`Slide1-1-scaled.png` to `Slide3-1-scaled.png`) are charts, not staged; its numbers are minimum-wage design apart from the £1,040. - - No Budget response briefing, chart pack or Top of the Charts Budget special from 28–29 October appeared. - - The Spread thin page's "Further reading" links two pre-Budget briefings, outside this window and not staged: +- **Publications we checked:** + - The lane fetched RF's publications, press-release, comment, Top of the Charts, events and Budgets listings at 16:19–16:20Z. + - RF put out one other Budget output on 28–29 October, the launch event page, *Pleasing people and satisfying markets*. RF embeds a YouTube video there and prints no figures, so we neither hash nor stage it. + - RF illustrates the comment with three slide images (`Slide1-1-scaled.png` to `Slide3-1-scaled.png`), and we didn't stage those charts. Apart from the £1,040, RF uses numbers in the comment only for minimum-wage design. + - We found no Budget response briefing, chart pack or Top of the Charts Budget special from 28–29 October. + - Under "Further reading" on the Spread thin page, RF links two pre-Budget briefings, which fall outside this window and which we didn't stage: - *Counting the cost* (8 Oct; page sha256 `415deda6…`). - *Hitting the buffers* (13 Oct; `881bbd4d…`). ## Discrepancies in the primaries (reported, not fixed) -1. **v1 and v2 disagree on the income halves.** v1 prints £158 / £14 / £144; v2 prints £145 / £15 / £130, and v2 p.19 hedges them as "under £145 … over £130 … less than £15". The landing page still printed £158 / £144 when fetched on 29 October; by 4 November it printed v2's £145 / £130 (manifest line 8). -2. **The £4.4bn package's years.** p.19 calls it "the Government’s £4.4 billion cost of living package for 2025-26 and 2026-27". The measures fall in 2026-27 and 2027-28 (Table 6.1 lines 1–3). The press release and p.47 print no years. The family stages the period as printed, at medium confidence. -3. **The £3.4bn's scope.** p.17 says "Together, these measures cost £3.4 billion in 2026-27" of the Household Energy Payment and "a three-month extension to the VAT cut". The extension falls in 2027-28, and £3.4bn equals lines 1 and 2 in 2026-27 (£2,625m + £755m), so the figure includes the original six months of the VAT cut. -4. **Your First Home's sign.** p.21 calls £95m in 2029-30 "a cost". Table 6.1 line 37 prints +95, a yield ("net income from developer fees"). -5. **SAHP top-up** (not staged, rule 8). p.21's "£1.5 billion top-up" differs from Table 6.1 line 35, which sums to £1,750m over 2027-28 to 2030-31. -6. **Good Growth Funds** (not staged, rule 8). p.33's "£285 million in 2027-28" differs from line 20, which prints -£340m that year (the line includes the lending subsidy). -7. **Tax gap measures.** p.4's "over £3 billion annually by 2030-31" doesn't match the workbook's OBR series (F10 row 52): 2.89 in 2030-31 and 3.64 in 2031-32. Row 67 records it in `conditions.discrepancy` (review A, K-7(h)). -8. **Figure 20's AB26 component.** The note says "‘AB26 Changes’ includes the Pension Tax Allowance, removal of standing charges on gas bills, and the Machine Games Levy". The workbook's AB26 Changes also includes "Property and savings tax rise" (column K = F + G + I + J). The workbook labels the Pension Tax Allowance "Pensioner Personal Allowance". -9. **Figure 10's Fuel Duty series.** Its note says the series shows "Fuel Duty changes announced in May and October", but the values (-0.575, -0.78) equal Table 6.1 line 3 alone. -10. **Primary balance years.** p.39 says the £107bn of higher tax revenues runs to the end of the Parliament (2028-29). Footnote 53 computes it "between 2023-24 and 2029-30, multiplied by forecasted GDP in 2029-30". +1. **v1 and v2 disagree on the income halves.** RF prints £158 / £14 / £144 in v1 and £145 / £15 / £130 in v2, and on v2 p.19 hedges them as "under £145 … over £130 … less than £15". RF still printed £158 / £144 on the landing page when we fetched it on 29 October; by 4 November it printed v2's £145 / £130 (manifest line 8). +2. **The £4.4bn package's years.** On p.19 RF calls it "the Government’s £4.4 billion cost of living package for 2025-26 and 2026-27". The measures fall in 2026-27 and 2027-28 (Table 6.1 lines 1–3), and RF prints no years in the press release or on p.47, so we stage the period as printed, at medium confidence. +3. **The £3.4bn's scope.** On p.17 RF says "Together, these measures cost £3.4 billion in 2026-27" of the Household Energy Payment and "a three-month extension to the VAT cut". The extension falls in 2027-28, and £3.4bn equals lines 1 and 2 in 2026-27 (£2,625m + £755m), so RF's figure includes the original six months of the VAT cut. +4. **Your First Home's sign.** On p.21 RF calls £95m in 2029-30 "a cost". HM Treasury prints +95 on Table 6.1 line 37, a yield ("net income from developer fees"). +5. **SAHP top-up** (not staged, rule 8). RF's "£1.5 billion top-up" on p.21 differs from Table 6.1 line 35, which sums to £1,750m over 2027-28 to 2030-31. +6. **Good Growth Funds** (not staged, rule 8). RF's "£285 million in 2027-28" on p.33 differs from line 20, on which HM Treasury prints -£340m that year (the line includes the lending subsidy). +7. **Tax gap measures.** RF's "over £3 billion annually by 2030-31" on p.4 doesn't match the workbook's OBR series (F10 row 52): 2.89 in 2030-31 and 3.64 in 2031-32. We record it in row 67's `conditions.discrepancy` (review A, K-7(h)). +8. **Figure 20's AB26 component.** In its note RF writes "‘AB26 Changes’ includes the Pension Tax Allowance, removal of standing charges on gas bills, and the Machine Games Levy". In the workbook RF's AB26 Changes also includes "Property and savings tax rise" (column K = F + G + I + J). RF labels the Pension Tax Allowance "Pensioner Personal Allowance" in the workbook. +9. **Figure 10's Fuel Duty series.** In its note RF describes the series as "Fuel Duty changes announced in May and October", but the values (-0.575, -0.78) equal Table 6.1 line 3 alone. +10. **Primary balance years.** On p.39 RF says the £107bn of higher tax revenues runs to the end of the Parliament (2028-29). In footnote 53 RF computes it "between 2023-24 and 2029-30, multiplied by forecasted GDP in 2029-30". 11. **Energy bill figures, release vs briefing:** - - The 28 October release's "£2,210 across 2027" (the 2027 average) differs from Spread thin's £2,192 (the Q1 2027 cap), which p.5 calls the bill "next year". + - The 28 October release's "£2,210 across 2027" (the 2027 average) differs from Spread thin's £2,192 (the Q1 2027 cap), which RF calls the bill "next year" on p.5. - The 29 October release's 2022-23 net bill of £1,685 differs from Figure 7's "After rebate: £1,686" (workbook 1,686.20). - p.17's VAT extension "worth £12" differs from the workbook's 12.80 for the cut in 2027-28 (2026-27 prices). -12. **Footnote 50's figure reference.** It reads "From the forecasts shown in Figure 7", but the current-balance forecasts appear in Figure 18; Figure 7 is the energy bill chart. +12. **Footnote 50's figure reference.** RF writes "From the forecasts shown in Figure 7", but it shows the current-balance forecasts in Figure 18; Figure 7 charts energy bills. 13. **Capacity vs Table 6.2** (not staged, rule 8). The NWF's £15.4bn and BBB's £5bn "capacity" boosts don't match Table 6.2's financial transactions: NWF line 2 sums to £10.2bn over 2028-29 to 2031-32, BBB line 3 to £0.75bn. -14. *(Withdrawn on 3 November: the README's rule 3 example was corrected, so RF's package equals the registered one; see "Measure keys".)* -15. **The VAT element's description.** PR28 para 5 and p.17's heading and paragraph 1 describe the package's VAT element as the "three-month extension", and p.17 calls the six-month cut "support announced before this Budget". Yet Figure 8, the 52/48 split and the £4.4bn all include "Energy VAT cut (original)". +14. *(Withdrawn on 3 November: the coordinator corrected the README's rule 3 example, so RF's package equals the registered one; see "Measure keys".)* +15. **The VAT element's description.** In PR28 para 5 and p.17's heading and paragraph 1 RF describes the package's VAT element as the "three-month extension", and on p.17 it calls the six-month cut "support announced before this Budget". Yet Figure 8, the 52/48 split and the £4.4bn all include "Energy VAT cut (original)". ## Review fixes, 3 November 2026 -The RF review (`harvest-staging/ab2026/reviews/verify-rf.md` on the staging branch) approved this family with fixes; all thirteen are applied, finding 1 as the `P_SAV_PROP` constant and the rest in an explicit block before `write()` in `stage.py`. No primary was re-opened (the session's network policy refused resolutionfoundation.org); no printed value changed. +In its review (`harvest-staging/ab2026/reviews/verify-rf.md` on the staging branch), the RF reviewer approved this family with fixes, and we applied all thirteen: finding 1 as the `P_SAV_PROP` constant, and the rest in an explicit block before `write()` in `stage.py`. We re-opened no primary, because the session's network policy refused resolutionfoundation.org, and changed no printed value. -1. The 25 rows for exactly lines 61 and 62 use `ab2026__package_property_and_savings_income_rates`, README rule 2's name (F10 row 51 equals lines 61 + 62 in every year). -2. The passage that contradicted README rule 3's old example is gone, and discrepancy 14 is withdrawn (kept as a one-line marker); that example has been corrected. -3. Row 168's note names the Government's Defence Investment Plan projection. -4. Row 38's note names the ONS/OBR outturn. -5. The v1→v2 record notes Figure 8's lower Council Tax rebate bars. -6. "Figure 7's four annotations" now reads two, with the two "Net bill" labels staged from p.17. -7. The workbook hash typo is fixed. -8. Row 23 (−0.4 per cent) is a `point`, `statistic: RHDI per person, average annual growth`, `parse_confidence: medium`. -9. Row 167 (0.5 per cent) is `statistic: real RDEL per person, average annual change (Figure 14)`. -10. Rows 28, 29, 56, 57, 61 and 63 carry `price_basis: cash terms (Figure 8)`. -11. (Optional, applied.) Row 397, F10 C48's floating-point zero (−7.46e-7 bn), is dropped: F10 54 rows, `ab2026_option__fuel_duty_continued_freeze` 6, `arithmetic` 148, total 766. -12. (Optional, applied; England, the coordinator's call.) Rows 109 and 110 (£358 and 20 per cent) are `geography: England`, as row 108 is. -13. Discrepancy 15 records the VAT element's description. +1. For the 25 rows for exactly lines 61 and 62 we use `ab2026__package_property_and_savings_income_rates`, README rule 2's name (F10 row 51 equals lines 61 + 62 in every year). +2. We removed the passage that contradicted README rule 3's old example, and withdrew discrepancy 14 (kept as a one-line marker); the coordinator has corrected that example. +3. In row 168's note we name the Government's Defence Investment Plan projection. +4. In row 38's note we name the ONS/OBR outturn. +5. In the v1→v2 record we note Figure 8's lower Council Tax rebate bars. +6. We changed "Figure 7's four annotations" to two, and stage the two "Net bill" labels from p.17. +7. We fixed the workbook hash typo. +8. We stage row 23 (−0.4 per cent) as a `point`, `statistic: RHDI per person, average annual growth`, `parse_confidence: medium`. +9. We give row 167 (0.5 per cent) `statistic: real RDEL per person, average annual change (Figure 14)`. +10. We give rows 28, 29, 56, 57, 61 and 63 `price_basis: cash terms (Figure 8)`. +11. (Optional, applied.) We dropped row 397, F10 C48's floating-point zero (−7.46e-7 bn): F10 54 rows, `ab2026_option__fuel_duty_continued_freeze` 6, `arithmetic` 148, total 766. +12. (Optional, applied; England, the coordinator's call.) We set rows 109 and 110 (£358 and 20 per cent) to `geography: England`, as on row 108. +13. We record the VAT element's description as discrepancy 15. -- **Verification:** the contract checker reports 766 rows, 0 problems, and a re-run of `stage.py` reproduces the gzip byte for byte. COUNTS.json pins its sha256. Proposed-key names may change again in the cross-family harmonisation recorded below. +- The contract checker reports 766 rows, 0 problems, and when we re-run `stage.py` it reproduces the gzip byte for byte. We pin its sha256 in COUNTS.json. The coordinator may change proposed-key names again in the cross-family harmonisation recorded below. ## Key harmonisation, 3 November 2026 -The coordinator gave each scope one key across families (tranche 1, task 3.4). This family's renames: +The coordinator gave each scope one key across families (tranche 1, task 3.4), and renamed these keys in this family: - `ab2026__minimum_wage_rates_april_2027` → `ab2026__minimum_wage_uprating_april_2027` - `ab2026__package_obr_total_effect_of_government_decisions` → `ab2026__package_government_decisions_since_march_2026` @@ -387,37 +393,37 @@ The coordinator gave each scope one key across families (tranche 1, task 3.4). T - `ab2026__package_obr_tax_administration_compliance_and_debt_collection` → `ab2026__package_tax_administration_compliance_and_debt_collection` - `ab2026_option__obr_tax_measures_with_continued_fuel_duty_freeze` → `ab2026_option__tax_decisions_with_continued_fuel_duty_freeze` -RF's three OBR groupings are restatements of the OBR's own Table 3.1 rows, so they take the OBR family's scope names, which carry no `obr` (OBR review, finding 9). The minimum wage key takes the red book's spelling for the same measure. `ab2026_option__fuel_duty_continued_freeze` is now shared with the OBR family, whose paragraph 3.43 prints the same alternative's £4.3bn in 2029-30. +In its three OBR groupings RF restates the OBR's own Table 3.1 rows, so we give them the OBR family's scope names, which drop `obr` (OBR review, finding 9). For the minimum wage key we take the red book's spelling for the same measure. We now share `ab2026_option__fuel_duty_continued_freeze` with the OBR family, where the OBR prints the same alternative's £4.3bn in 2029-30 in paragraph 3.43. ## README rule 10, 3 November 2026 -Under README rule 10, `stage.py` moves the family's totals of the Budget's decisions from `revenue_change` to `decisions_effect_on_borrowing`: Table 6.1's total of policy decisions (1 row) and the OBR's direct effect of tax measures that Figure 10 plots (13 rows), 14 rows in all. Counting from 0, the rows are 66, 102, 141–146 and 385–390. The OBR's total effect of the Budget's decisions (3 rows) already took the metric, so 17 rows carry it. Single measures and groups of measures keep `revenue_change`. No value or sign changed. +Under README rule 10, in `stage.py` we move the family's totals of the Budget's decisions from `revenue_change` to `decisions_effect_on_borrowing`. We move Table 6.1's total of policy decisions (1 row) and the OBR's direct effect of tax measures that Figure 10 plots (13 rows), 14 rows in all. Counting from 0, the rows are 66, 102, 141–146 and 385–390. We had already given the OBR's total effect of the Budget's decisions (3 rows) that metric, so we use it on 17 rows. We keep `revenue_change` for single measures and groups of measures, and changed no value or sign. -**Not moved (4 November 2026, review B's RF finding 3, option (b), the coordinator's call):** the 11 rows of Figure 10's "Net effect if Fuel Duty frozen" (`ab2026_option__tax_decisions_with_continued_fuel_duty_freeze`; rows 147–151 and 432–437), which the 3 November pass had moved, are back on `revenue_change`. These totals add the freeze's receipts effect, which row 154's note derives from the OBR's constant-duty-rate Fuel Duty forecast less its central forecast, and RF's title for Figure 10 frames the chart as revenue ("Measures announced raise revenues"). Under rule 10 they are a tax-change estimate whose evidence doesn't establish PSNB scope. The OBR states the freeze's effect on borrowing for 2029-30 only (paragraph 3.43, from Table 7.1), and RF's values for the other years are receipts differences, so they are not recombinations of a stated effect on borrowing. Rows 154 and 397–401 (`ab2026_option__fuel_duty_continued_freeze`) stay on `revenue_change` too. +**Not moved (4 November 2026, review B's RF finding 3, option (b), the coordinator's call).** The 3 November pass had moved the 11 rows of Figure 10's "Net effect if Fuel Duty frozen" (`ab2026_option__tax_decisions_with_continued_fuel_duty_freeze`; rows 147–151 and 432–437). We put them back on `revenue_change`. RF's totals here add the freeze's receipts effect, which we derive in row 154's note from the OBR's constant-duty-rate Fuel Duty forecast less its central forecast. In its title for Figure 10 RF also frames the chart as revenue ("Measures announced raise revenues"). Under rule 10 we treat them as a tax-change estimate whose evidence doesn't establish PSNB scope. The OBR states the freeze's effect on borrowing for 2029-30 only (paragraph 3.43, from Table 7.1). For the other years RF gives receipts differences, so they don't recombine a stated effect on borrowing. We keep rows 154 and 397–401 (`ab2026_option__fuel_duty_continued_freeze`) on `revenue_change` too. ## Review fixes, 4 November 2026 -Two independent reviews of PR #231 at 4f424b9 approved this family with fixes: review B (RF findings 1–8 and nits) and review A (K-4, K-6, K-7(c), K-7(d), K-7(h) and H-1). All are applied in a block before `write()` in `stage.py`, with the coordinator's calls on findings 2 and 3. No printed or stored value changed. Indices are 4f424b9's. +Two independent reviews of PR #231 at 4f424b9 approved this family with fixes: review B (RF findings 1–8 and nits) and review A (K-4, K-6, K-7(c), K-7(d), K-7(h) and H-1). We applied all of them in a block before `write()` in `stage.py`, with the coordinator's calls on findings 2 and 3, and changed no printed or stored value. We give 4f424b9's indices. -1. **Two-year totals (finding 1).** All 114 Figure 8 rows (245–358), the text rows that restate them (56, 57, 61, 63, 74, 76), the gap rows 62 and 75 and the £4.4bn rows 34, 54 and 127 are `cumulative`. The 95 F8 £ rows' normalization reads "£ per household as stored: total cash impact over 2026-27 and 2027-28 (Figure 8 note)"; the K rows keep theirs. -2. **The landing page changed by 4 November (finding 2, rule 7).** Manifest line 8 records the 4 November page (no rows cite it). Its £145 / £130 are not staged again (the coordinator's call; v2 rows 74 and 76 carry them). Rows 56–57 carry `document_version: "29 October page text (superseded by 4 November)"` and a "Superseded" sentence. -3. **Rule 10 (finding 3, option (b)).** Rows 147–151 and 432–437 are back on `revenue_change`; the stager's rule 10 count is 14 (see "README rule 10"). -4. **Restated counts (finding 4).** Rows 55 and 72 are `restated`, `hm_government_publication`, medium confidence, with a note naming HM Treasury. -5. **Tallies (finding 5).** The workbook has 554 rows, the restated source models are corrected, and the attribution split recounts from the rows (`own` 680, `restated` 86). -6. **Gas standing charge (finding 6).** Rows 115–125 carry `conditions.horizon` and medium confidence. -7. **Signs (finding 7, and K-7(d)).** The 122 Figure 21–22 rows (629–750) carry "positive = growth"; text rows 21, 23, 89, 117, 198, 200, 203 and 204 carry "positive = gain to households"; row 10 reads "positive = saving (a reduction in the household's bills)". +1. **Two-year totals (finding 1).** We mark all 114 Figure 8 rows (245–358), the text rows that restate them (56, 57, 61, 63, 74, 76), the gap rows 62 and 75 and the £4.4bn rows 34, 54 and 127 `cumulative`. For the 95 F8 £ rows, we record "£ per household as stored: total cash impact over 2026-27 and 2027-28 (Figure 8 note)" in `normalization`, and keep each K row's own normalization. +2. **RF changed the landing page by 4 November (finding 2, rule 7).** We record the 4 November page as manifest line 8 (we cite it in no row). The coordinator chose not to stage its £145 / £130 again, since we stage them in v2 rows 74 and 76. We set `document_version: "29 October page text (superseded by 4 November)"` and a "Superseded" sentence on rows 56–57. +3. **Rule 10 (finding 3, option (b)).** We put rows 147–151 and 432–437 back on `revenue_change`, so the stager counts 14 under rule 10 (see "README rule 10"). +4. **Restated counts (finding 4).** We mark rows 55 and 72 `restated`, `hm_government_publication`, medium confidence, and name HM Treasury in each note. +5. **Tallies (finding 5).** We count 554 workbook rows, corrected the restated source models and recounted the attribution split from the rows (`own` 680, `restated` 86). +6. **Gas standing charge (finding 6).** We gave rows 115–125 `conditions.horizon` and medium confidence. +7. **Signs (finding 7, and K-7(d)).** We give the 122 Figure 21–22 rows (629–750) "positive = growth", and text rows 21, 23, 89, 117, 198, 200, 203 and 204 "positive = gain to households". We give row 10 "positive = saving (a reduction in the household's bills)". 8. **Nits (finding 8):** - - chart rows name their series in `conditions.series` (310 rows), and `component` stays only on the 39 rule 9 parts of line 2 (review A, H-1). Seven text context rows that used `component` for a driver of a forecast change, or a scenario's spending type, use `driver` (rows 51, 80–82, 91, 92) and `spending_type` (row 170); - - F22's category labels read "2025-26", not "2025- 26"; - - stored F7 C/I/M and F9 cells say "as stored"; - - F7 rows 222, 226, 236 and 241 name their support series in `reform_hint`; - - gap rows 30, 62 and 75 carry `price_basis: cash terms (Figure 8)`; - - the Primaries notes, Figure 21's chart object, F23's second chart, F8 J, discrepancy 15's locator, Review fixes item 2 and the "Not staged" list are corrected. -9. **K-4.** The £4.4bn (34, 54, 127) moves to `ab2026__package_household_energy_vat_and_household_share_of_fuel_duty`. -10. **K-6.** `ab2026_option__fuel_duty_rpi_uprating_since_2011` stays separate from the OBR's `ab2026__package_fuel_duty_freezes_since_2010_11`; each family's NOTES names the other. -11. **K-7(c).** Rows 88 and 211 propose `gross_earnings_change`. -12. **K-7(h).** Row 67 carries `conditions.discrepancy`. -13. **OBR review B, finding 3.** The tax administration group's key is `ab2026__package_tax_administration_compliance_and_debt_collection_receipts`, as the OBR family's is. - -- **For tranche 2:** `ab2026_option__fuel_duty_continued_freeze` holds RF's receipts cost (rows 154, 397–401, `revenue_change`) and the OBR's £4.3bn effect on borrowing (paragraph 3.43). RF rows 385–390 restate the OBR workbook's receipts total exactly, on the tax-decisions key with the borrowing metric. -- **Verification:** the contract checker reports 766 rows, 0 problems, and a re-run of `stage.py` in a scratch directory reproduces the gzip byte for byte. + - We name each chart row's series in `conditions.series` (310 rows), and keep `component` only on the 39 rule 9 parts of line 2 (review A, H-1). On seven text context rows we had used `component` for a driver of a forecast change, or a scenario's spending type; we now use `driver` (rows 51, 80–82, 91, 92) and `spending_type` (row 170); + - we write F22's category labels "2025-26", not "2025- 26"; + - for stored F7 C/I/M and F9 cells we say "as stored"; + - we name the support series of F7 rows 222, 226, 236 and 241 in `reform_hint`; + - we gave gap rows 30, 62 and 75 `price_basis: cash terms (Figure 8)`; + - we corrected the Primaries notes, Figure 21's chart object, F23's second chart, F8 J, discrepancy 15's locator, Review fixes item 2 and the "Not staged" list. +9. **K-4.** We moved the £4.4bn (34, 54, 127) to `ab2026__package_household_energy_vat_and_household_share_of_fuel_duty`. +10. **K-6.** We keep `ab2026_option__fuel_duty_rpi_uprating_since_2011` separate from the OBR's `ab2026__package_fuel_duty_freezes_since_2010_11`, and name each family in the other's NOTES. +11. **K-7(c).** We propose `gross_earnings_change` on rows 88 and 211. +12. **K-7(h).** We set `conditions.discrepancy` on row 67. +13. **OBR review B, finding 3.** We key the tax administration group to `ab2026__package_tax_administration_compliance_and_debt_collection_receipts`, as in the OBR family. + +- We key both RF's receipts cost (rows 154, 397–401, `revenue_change`) and the OBR's £4.3bn effect on borrowing (paragraph 3.43, `uk_obr_tables` row 2740) to `ab2026_option__fuel_duty_continued_freeze`. At ingest we drop RF's six rows as restatements of the OBR's forecast, and keep the OBR's row on `decisions_effect_on_borrowing` under the macro verdict (#55), because the OBR states its effect on public sector net borrowing. In rows 385–390 RF restates the OBR workbook's receipts total exactly; we key them to the tax-decisions key with the borrowing metric, and at ingest we drop them as restatements. +- The contract checker reports 766 rows, 0 problems, and when we re-run `stage.py` in a scratch directory it reproduces the gzip byte for byte. diff --git a/sources/harvest-uk-ab2026-2026-10-29/uk_thinktank_misc/NOTES.md b/sources/harvest-uk-ab2026-2026-10-29/uk_thinktank_misc/NOTES.md index cb9246af8..0af0df50e 100644 --- a/sources/harvest-uk-ab2026-2026-10-29/uk_thinktank_misc/NOTES.md +++ b/sources/harvest-uk-ab2026-2026-10-29/uk_thinktank_misc/NOTES.md @@ -1,31 +1,31 @@ # Smaller producers: Autumn Budget 2026 -**Working files.** `stage.py`, which writes `manifest.jsonl` and `claims_staged.jsonl.gz`, lives on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/thinktanks-b/uk_thinktank_misc/stage.py`; this branch carries only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz`. The lane's local primaries folder (`.ab2026-primaries/`) was never committed: the repository carries no primary bytes, so fetch each primary from its `manifest.jsonl` URL and check its sha256. +We keep `stage.py`, which writes `manifest.jsonl` and `claims_staged.jsonl.gz`, on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/thinktanks-b/uk_thinktank_misc/stage.py`. We commit only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz` to this branch, and no primary bytes to the repository. The lane never committed its local primaries folder (`.ab2026-primaries/`), so fetch each primary from its `manifest.jsonl` URL and check its sha256. -Staged from five primary HTML responses published on 28 October 2026. Sources are `cps`, `institute_for_government`, `learning_and_work_institute` and `citizens_advice`; the manifested Age UK response contributes zero rows. There are **12 rows and 5 primaries**. `stage.py` embeds the readings and recreates the manifest and deterministic gzip using only Python's standard library. +We staged this family from five primary HTML responses that their producers published on 28 October 2026. We use the sources `cps`, `institute_for_government`, `learning_and_work_institute` and `citizens_advice`; we manifested Age UK's response but staged zero rows from it. We stage **12 rows and 5 primaries**. We embedded the readings in `stage.py`, which recreates the manifest and deterministic gzip with only Python's standard library. ## Changes on 3 November 2026 (thinktanks-b review) -- **Finding 1:** CPS's £3.1 billion PuFins row is context. It names no Budget decision, line or year and ties the figure to a CPS paper from before the Budget, so it carries no proposed key; `reform_hint` is null and `conditions.measure_type` names it. `ab2026__package_cps_public_financial_institutions` is withdrawn. -- **Finding 5:** Trussell's row and primary moved to the AB2026 family `uk_trussell_wpi`, with source `trussell_wpi`, as AB2025 staged Trussell's Budget responses. This family now has 12 rows and 5 primaries. -- **Finding 6:** IfG's count of one tax measure names the measure it counted in `conditions.subgroup`, not `component`, which rule 9 reserves for parts of a measure. +- Finding 1: we made CPS's £3.1 billion PuFins row context. CPS names no Budget decision, line or year and ties the figure to a CPS paper from before the Budget, so we gave the row no proposed key. We set `reform_hint` to null, named the row's subject in `conditions.measure_type` and withdrew `ab2026__package_cps_public_financial_institutions`. +- Finding 5: we moved Trussell's row and primary to the AB2026 family `uk_trussell_wpi`, with source `trussell_wpi`, as we staged Trussell's Budget responses in AB2025. This family now has 12 rows and 5 primaries. +- Finding 6: for IfG's count of one tax measure, we now name the measure it counted in `conditions.subgroup`, not `component`, which we use only for parts of a measure (rule 9). -No primary was re-opened on 3 November (the session's network policy refused every producer's site); every change relabels fields, and no printed value changed. The 29 October record below is updated for the new row set. +We re-opened no primary on 3 November, because the session's network policy refused every producer's site. We only relabelled fields, changed no printed value and updated the 29 October record below for the new row set. ## Changes on 4 November 2026 (reviews A and B) -Review B (PR #231, issue comment 6635498758) approved this family with fixes, and review A's H-6(a) asked for its finding 2. The fixes sit in a block after the rows in `stage.py`, shared word for word with `uk_trussell_wpi`'s (judgement call 10). No primary was re-opened, and no value changed. +Review B (PR #231, issue comment 6635498758) approved this family with fixes, and review A's H-6(a) asked for its finding 2. We put the fixes in a block after the rows in `stage.py`, shared word for word with `uk_trussell_wpi`'s (judgement call 10). We re-opened no primary and changed no value. -- **Finding 1:** IfG's "around 80%" (row 8) restates HM Treasury's paragraph 3.5, which `uk_hmt_redbook` row 191 stages: `restated`, `source_model: administrative_data`, `conditions.originator: "HM Treasury, Budget 2026 para 3.5"`, geography England, denominator "households in England (HM Treasury)", and a rewritten note. Own 4, restated 8. -- **Finding 2 and H-6(a):** rows 2, 8 and 10 carry the registered `fiscal_headroom`, `affected_share` and `affected_count` in `metric`, with `proposed_metric` null. -- **Finding 3:** row 9's `measure_type` is "real household disposable income per person". -- **H-6(d):** row 11 (Citizens Advice) carries `geography: "not stated"`. +- Finding 1: in "around 80%" (row 8) IfG restates HM Treasury's paragraph 3.5, which we stage as `uk_hmt_redbook` row 191. We now mark the row `restated`, with `source_model: administrative_data`, `conditions.originator: "HM Treasury, Budget 2026 para 3.5"`, geography England, denominator "households in England (HM Treasury)" and a rewritten note. Own 4, restated 8. +- Finding 2 and H-6(a): we now record the registered `fiscal_headroom`, `affected_share` and `affected_count` in `metric` on rows 2, 8 and 10, with `proposed_metric` null. +- Finding 3: we set row 9's `measure_type` to "real household disposable income per person". +- H-6(d): we set `geography: "not stated"` on row 11 (Citizens Advice). ## Access recipe and primary hashes -Every primary was fetched afresh on 29 October 2026 using `curl -sSL -A "Mozilla/5.0 (Macintosh; Intel Mac OS X 14_0) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/17.0 Safari/605.1.15"`; every response was HTTP 200. The lane kept the HTML bytes, headers and text extracts in its local primaries folder. Hashes were checked using `shasum -a 256`. BeautifulSoup extracts were used to read the page's own body text; hashes refer to HTML, not extracts. There are no PDFs or spreadsheets in this family. +We fetched every primary afresh on 29 October 2026 with `curl -sSL -A "Mozilla/5.0 (Macintosh; Intel Mac OS X 14_0) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/17.0 Safari/605.1.15"`, and every server answered HTTP 200. The lane kept the HTML bytes, headers and text extracts in its local primaries folder and checked hashes with `shasum -a 256`. It read each page's own body text from BeautifulSoup extracts, but hashed the HTML, not the extracts. We staged no PDF or spreadsheet in this family. -No assigned local copies or earlier document versions were supplied. There was no blocked fetch, fallback or observed replacement. Fetch-completion timestamps are recorded in each manifest note because HTML can change. +Nobody supplied assigned local copies or earlier document versions, and we hit no blocked fetch, used no fallback and saw no replacement. Because HTML can change, we record the fetch-completion time in each manifest note. | Primary | Fetch completion (UTC) | Bytes | Rows | SHA256 | |---|---|---:|---:|---| @@ -35,11 +35,11 @@ No assigned local copies or earlier document versions were supplied. There was n | [Learning and Work Institute response](https://learningandwork.org.uk/responding-to-autumn-budget-2026/) | 2026-10-29 15:22:31 | 122,809 | 1 | `c7fe077a0d9f4784aa4c91c906bcefa087cc9a4ece8b28026d24347c2a8b5dcf` | | [Citizens Advice response](https://www.citizensadvice.org.uk/about-us/media-centre/press-releases/citizens-advice-responds-to-chancellors-budget/) | 2026-10-29 15:23:24 | 22,671 | 1 | `6a10336a67fc058512a4ccd9d0e80a27470fee8ba5775690fb1fcba39db67385` | -Citizens Advice's press index linked its dated response. The other four URLs were assigned. CPS prints “Date Added: Wednesday 28th October 2026”; IfG prints “28 OCT 2026”; LWI prints “28 October 2026”; Age UK prints publication on 28 October 2026 at 04:14 PM and last update that date. +We followed Citizens Advice's press index to its dated response; the coordinator assigned the other four URLs in the lane's brief (`harvest-staging/ab2026/briefs/thinktanks-b.md` on the staging branch). CPS prints “Date Added: Wednesday 28th October 2026”, IfG prints “28 OCT 2026” and LWI prints “28 October 2026”. Age UK prints publication on 28 October 2026 at 04:14 PM and a last update that date. ## Coverage and conventions -Reading order is CPS, IfG, Age UK, LWI, Citizens Advice, then top to bottom within each page. Only contiguous excerpts under 15 words are quoted. There are no chart readings or derived gaps, totals, ratios or annualisations. +We read in the order CPS, IfG, Age UK, LWI, Citizens Advice, then top to bottom within each page, and quote only contiguous excerpts under 15 words. We read no charts and derived no gaps, totals, ratios or annualisations. | Source | Rows | Own | Restated | |---|---:|---:|---:| @@ -53,56 +53,56 @@ Reading order is CPS, IfG, Age UK, LWI, Citizens Advice, then top to bottom with | Measure key | Rows | Scope | |---|---:|---| | `ab2026__household_energy_payment_100_bands_a_to_d` | 1 | IfG's share of households in England receiving the £100 council-tax rebate for bands A–D (restating HM Treasury's paragraph 3.5); Table 6.1 line 1 | -| `null` | 11 | context and document-tally observations; none carries a proposed key | +| `null` | 11 | context and document-tally observations; we proposed a key for none | -The sole registered key is attached to the specific rebate's coverage, not to the entire cost-of-living package. IfG's exact identification of the £100 council-tax rebate for bands A–D identifies line 1; it gives no contrary scope. Its "around 80%" restates HM Treasury's paragraph 3.5 ("around 80% of households in England"), so the row is `restated`, `administrative_data`, geography England, as `uk_hmt_redbook` row 191 stages it (changed on 4 November, review B finding 1). Context forecasts, outturns and document counts, including CPS's PuFins assertion, have null measure keys and null reform hints. +We attach the sole registered key to the specific rebate's coverage, not to the entire cost-of-living package. IfG names the £100 council-tax rebate for bands A–D, so we match it to line 1, and IfG gives no contrary scope. Its "around 80%" restates HM Treasury's paragraph 3.5 ("around 80% of households in England"). So we mark the row `restated`, `administrative_data`, geography England, as we stage `uk_hmt_redbook` row 191; we made that change on 4 November, after review B finding 1. We gave null measure keys and null reform hints to context forecasts, outturns and document counts, including CPS's PuFins assertion. -- GBP values preserve printed signs and cash magnitudes. “Tops £3 trillion” is the printed strict lower bound of debt stock. The six-month £61 billion debt-interest total is not multiplied to a year. Fiscal headroom is positive headroom; PuFins' £3.1 billion is a positive debt increase. No sign is reversed. -- Percentages remain percentages. “Almost 100” retains the printed upper threshold 100; no exact policy count is inferred. “Around 80%” retains the qualifier in conditions. -- `point`, rather than `central`, records each single published observation/forecast: none is labelled a central scenario. Thresholds use `range_low` or `range_high`. The 2% real-income growth is `cumulative` over the printed comparison window, not a 2% annual rate. -- Only IfG's fiscal-rule probabilities name a fiscal year: “2029/30” is normalised to `fy: 2029-30`, period 2030, `fiscal_year`. The income projection prints calendar 2030 and uses `annual` without `fy`. Other observations use the publication-year point-in-time anchor 2026 and preserve “last six months”, “this winter”, “each year”, “last 12 months”, or unspecified horizon explicitly. They do not claim a fiscal/calendar scoring year that the producer did not print. The PuFins annual-rate assertion has medium confidence because no scoring year or fiscal/calendar basis is supplied. -- Real income remains real; all other price bases are left unspecified where the page does not identify one. No baseline proposal is needed for these current observations and explicitly named context comparisons. -- CPS's own PuFins assertion uses `source_model: unstated` because it does not identify its calculation. Official debt/outturn context uses `administrative_data` and `administrative_fact`. CPS headroom and IfG fiscal probabilities/income forecasts are `restated`, `obr_efo_forecast`, `different_model`. The precise original release behind the unsourced CPS figures is not identified, so these originator assignments are explicit in rows. -- IfG's two counts are its own `arithmetic`, `different_model` observations. Its rebate share restates HM Treasury (paragraph 3.5): `restated`, `administrative_data`, `conditions.originator` naming the paragraph (changed on 4 November). LWI's rounded NEET figure is treated as restated official labour-market context, `administrative_data`, `administrative_fact`; its response gives neither a quarter, age range nor a citation. Citizens Advice's unsourced bill forecast is treated as its own assertion with `source_model: unstated`, with missing method/geography acknowledged. Citizens Advice's estimate, CPS's PuFins assertion and IfG's rebate share have medium confidence; the other nine rows have high confidence. +- We keep printed signs and cash magnitudes in GBP values and take “Tops £3 trillion” as the printed strict lower bound of debt stock; we don't multiply the six-month £61 billion debt-interest total to a year. We record fiscal headroom as positive headroom and PuFins' £3.1 billion as a positive debt increase, and we reversed no sign. +- We keep percentages as percentages. For “Almost 100” we keep the printed upper threshold 100 and infer no exact policy count. For “Around 80%” we keep the qualifier in conditions. +- We record each single published observation or forecast as `point`, not `central`, because no producer labels one a central scenario. We use `range_low` or `range_high` for thresholds. We record the 2% real-income growth as `cumulative` over the printed comparison window, not a 2% annual rate. +- Only IfG names a fiscal year, for its fiscal-rule probabilities: we normalise “2029/30” to `fy: 2029-30`, period 2030, `fiscal_year`. IfG prints calendar 2030 for the income projection, so we use `annual` without `fy`. For the other observations we use the publication-year point-in-time anchor 2026 and keep “last six months”, “this winter”, “each year”, “last 12 months” or the unspecified horizon explicitly. We claim no fiscal or calendar scoring year that the producer did not print. We give the PuFins annual-rate assertion medium confidence, because CPS supplies no scoring year and no fiscal or calendar basis. +- We keep real income real, and leave every other price basis unspecified where the producer names none. These current observations and explicitly named context comparisons need no baseline proposal. +- CPS doesn't identify the calculation behind its own PuFins assertion, so we use `source_model: unstated`. We use `administrative_data` and `administrative_fact` for official debt and outturn context. We mark CPS's headroom and IfG's fiscal probabilities and income forecasts `restated`, `obr_efo_forecast`, `different_model`. We couldn't identify the original release behind CPS's unsourced figures, so we state these originator assignments in the rows. +- We record IfG's two counts as its own `arithmetic`, `different_model` observations. In its rebate share IfG restates HM Treasury (paragraph 3.5), so since 4 November we mark it `restated`, `administrative_data`, and name the paragraph in `conditions.originator`. We treat LWI's rounded NEET figure as restated official labour-market context, `administrative_data`, `administrative_fact`; LWI gives no quarter, age range or citation. We treat Citizens Advice's unsourced bill forecast as its own assertion with `source_model: unstated`, and we record that it gives no method or geography. We give Citizens Advice's estimate, CPS's PuFins assertion and IfG's rebate share medium confidence, and the other nine rows high confidence. ## Proposed keys, metrics and units -No measure key is proposed. CPS's PuFins row is context (see the 3 November changes above). +We propose no measure key. We made CPS's PuFins row context (see the 3 November changes above). -Registered metrics (since 4 November, review A's H-6(a) and review B's finding 2; `models.py` registers them, so they go in `metric`, not `proposed_metric`): +Since 4 November, after review A's H-6(a) and review B's finding 2, we record these registered metrics in `metric`, not `proposed_metric`, because `models.py` registers them: - `fiscal_headroom` (1; GBP): CPS's current post-Budget buffer against the fiscal rule; target year unspecified. - `affected_share` (1; percent): IfG's share of households in England receiving the rebate, restating HM Treasury. - `affected_count` (1; persons): LWI's rounded current count of young people not in education, employment or training. -Proposed metrics (counts and exact meaning): +We propose these metrics (counts and exact meaning): -- `public_debt_level` (1; GBP): CPS's current national-debt stock lower bound; no series definition is supplied. +- `public_debt_level` (1; GBP): CPS's current national-debt stock lower bound; CPS supplies no series definition. - `debt_interest_cost` (1; GBP): CPS's total debt interest over the last six months, not the change caused by a measure. - `public_debt_change` (1; GBP): CPS's asserted annual debt increase through PuFins, a context row; period limitations above. - `measure_count` (2; proposed unit `measures`; AB2025 family precedent): IfG's approximate full scorecard count and exact count of tax measures costing/saving more than £1bn in the last forecast year. - `fiscal_rule_probability` (2; percent): current and last-budget OBR probabilities of a current-budget surplus in 2029/30, as reported by IfG. - `average_energy_bill_change` (1; GBP/month; AB2025 IFS/RF proposal spelling): Citizens Advice's forecast monthly household energy-bill increase, excluding any assertion that this is a Budget measure effect. -Registered `real_income_growth` accounts for IfG's restated OBR 2% cumulative growth in real household disposable income per person (EFO para 1.2; `conditions.measure_type` says "per person" since 4 November, review B finding 3). Proposed unit `measures` is used only for the two counts. Other units are registered. There are no proposed baselines. +We record IfG's restatement of the OBR's 2% cumulative growth in real household disposable income per person (EFO para 1.2) under the registered `real_income_growth`. Since 4 November, after review B finding 3, we write "per person" in `conditions.measure_type`. We use the proposed unit `measures` only for the two counts, and registered units for the rest. We propose no baselines. -## Not staged and primary limitations +## What we left out, and the primaries' limits -- **Age UK: zero rows.** Its own response has no numeric Budget estimate. The exact “On Income Tax threshold.” subsection was read; it welcomes protecting low-income pensioners from cash tax demands but prints no number. Dates, contact numbers, navigation “Nearly 4,000 flu mortalities” and the unrelated £2 donation link are not Budget claims. -- IfG's £100 rebate is a bare policy parameter; its 80% coverage is staged separately. The £1bn cutoff defines its count of one tax measure and is preserved in conditions rather than misrepresented as that measure's costing. Other dates and the 1991 council-tax valuation year are context labels, not model outcomes. -- LWI mentions extra Mayoral apprenticeship funding and OBR employment effects but prints no amount or employment effect. Its “one million” NEET level is retained as context. Navigation cards and unrelated articles are not harvested. -- Citizens Advice gives no source, exact winter dates, geography or method for £60/month (`geography: "not stated"` since 4 November, for the contract's closed geography list). Those are explicit limitations, not supplied from secondary material. -- CPS does not link its “paper released last week”. It is outside the assigned 28–29 October post-Budget window, and no unseen figures from it are staged. The comparison between six-month debt interest and an annual defence budget prints no second amount or exact ratio, so none is derived. -- No restated figure is silently removed: the seven official/OBR context rows and IfG's restatement of HM Treasury's 80% are staged with `restated` and an originator. No PolicyEngine restatement was identified. -- No pages were unreached. No charts contain unstaged numeric labels in these primaries. +- **Age UK: zero rows.** Age UK prints no numeric Budget estimate in its own response. We read the exact “On Income Tax threshold.” subsection; Age UK welcomes protecting low-income pensioners from cash tax demands there but prints no number. We didn't treat dates, contact numbers, the navigation's “Nearly 4,000 flu mortalities” or the unrelated £2 donation link as Budget claims. +- IfG gives the £100 rebate as a bare policy parameter, and we stage its 80% coverage separately. IfG counts one tax measure above a £1bn cutoff, and we keep the cutoff in conditions, not as that measure's costing. We treat the other dates and the 1991 council-tax valuation year as context labels, not model outcomes. +- LWI mentions extra Mayoral apprenticeship funding and OBR employment effects but prints no amount or employment effect. We keep its “one million” NEET level as context, and we harvested no navigation cards or unrelated articles. +- Citizens Advice gives no source, exact winter dates, geography or method for £60/month. Since 4 November we set `geography: "not stated"`, to fit the contract's closed geography list. We record those gaps as explicit limitations and filled none of them from secondary material. +- CPS does not link its “paper released last week”. That paper falls outside the assigned 28–29 October post-Budget window, and we staged no unseen figures from it. CPS compares six-month debt interest with an annual defence budget but prints no second amount or exact ratio, so we derived none. +- We stage the seven official/OBR context rows and IfG's restatement of HM Treasury's 80% with `restated` and an originator. We found no PolicyEngine restatement. +- We reached every page, and found no chart in these primaries with unstaged numeric labels. -Potentially ambiguous source wording is retained: IfG describes the £100 rebate as reaching “around 80%” of households without naming the denominator's geography; HM Treasury's paragraph 3.5, which it restates, says households in England, so the row carries England (4 November). CPS's PuFins claim lacks components, annual dates and method, so it stays context; its £15 billion headroom lacks a target year. Age UK's Pension Tax Allowance paragraph contains no number despite the assignment's pointer. These gaps have not been repaired using other sources. +IfG describes the £100 rebate as reaching “around 80%” of households without naming the denominator's geography. In paragraph 3.5, which IfG restates, HM Treasury says households in England, so since 4 November we record England on the row. CPS gives no components, annual dates or method for its PuFins claim, so we keep that row as context. CPS also gives no target year for its £15 billion headroom. Age UK prints no number in its Pension Tax Allowance paragraph despite the assignment's pointer. We repaired none of these gaps from other sources. ## Verification -- **4 November:** the contract checker reports 12 rows, 0 problems, and a re-run of `stage.py` reproduces both files byte for byte. -- **3 November:** the contract checker reports 12 rows, 0 problems; a re-run of `stage.py` reproduces `manifest.jsonl` and `claims_staged.jsonl.gz` byte for byte (COUNTS.json pins the gzip's sha256). The fixes relabel fields only, so the lane's checks below stand for the rows that remain. -- **29 October (13 rows, including Trussell's, now in `uk_trussell_wpi`):** +- On 4 November, the contract checker reported 12 rows, 0 problems, and when we re-ran `stage.py` it reproduced both files byte for byte. +- On 3 November, the contract checker reported 12 rows, 0 problems, and when we re-ran `stage.py` it reproduced `manifest.jsonl` and `claims_staged.jsonl.gz` byte for byte (COUNTS.json pins the gzip's sha256). We only relabelled fields, so the lane's checks below stand for the rows that remain. +- On 29 October, the lane checked 13 rows, including Trussell's, now in `uk_trussell_wpi`: Command: @@ -112,9 +112,9 @@ PYTHONPATH=. uv run -q python pipeline/ab2026_harvest_contract.py sources/harves sources/harvest-uk-ab2026-2026-10-29/uk_thinktank_misc: 13 rows, 0 problems ``` -Re-ran `python3 sources/harvest-uk-ab2026-2026-10-29/uk_thinktank_misc/stage.py`; claims SHA256 before and after was **`d677dde1f96e629097696f5d17b575c8f1d34baaca42b96297cd46be773669ea`**. The gzip header has mtime zero and no original filename. Manifest byte counts and all six primary hashes were checked against saved HTML. All 13 quotes were verified as contiguous text in their pinned primary and under 15 words; each contains its `value_raw`. +The lane re-ran `python3 sources/harvest-uk-ab2026-2026-10-29/uk_thinktank_misc/stage.py`, and the claims SHA256 before and after read **`d677dde1f96e629097696f5d17b575c8f1d34baaca42b96297cd46be773669ea`**. The stager writes the gzip header with mtime zero and no original filename. The lane checked manifest byte counts and all six primary hashes, Trussell's included, against the saved HTML. It verified that all 13 quotes appear as contiguous text in their pinned primary, run under 15 words and contain their `value_raw`. -Random sample was `random.Random(230).sample(range(13), 10)`, using zero-based indexes and reporting one-based row numbers below. For each, the saved primary was reopened, its SHA256 checked, and the paragraph at the locator read to verify the number, sign, unit and period. The three remaining rows (CPS debt interest, IfG count of one, Trussell share) were also checked for quote/value correspondence and basis. +The lane drew its random sample with `random.Random(230).sample(range(13), 10)`, using zero-based indexes, and reports one-based row numbers below. For each, it reopened the saved primary, checked its SHA256 and read the paragraph at the locator to verify the number, sign, unit and period. It also checked the three remaining rows (CPS debt interest, IfG count of one, Trussell share) for quote/value correspondence and basis. | Sample row | Primary and locator | Rechecked value, sign, unit and period | |---:|---|---| @@ -129,4 +129,4 @@ Random sample was `random.Random(230).sample(range(13), 10)`, using zero-based i | 4 | CPS, Herring's PuFins paragraph | £3.1 billion positive debt increase each year → 3.1e9 GBP; no start/end year; 2026 statement anchor with annual horizon retained | | 11 | LWI, Evans's main quotation | “one million” young NEET people → 1e6 persons; current context at publication 2026; no quarter invented | -Scope review found one non-null key: IfG's specifically identified bands A–D £100 rebate, Table 6.1 line 1. No broader support-package figure is attached to that key. (On 3 November the CPS row lost its proposed key; see above.) No git state-changing commands, commits or pushes were run. +In its scope review the lane found one non-null key: IfG's specifically identified bands A–D £100 rebate, Table 6.1 line 1. We attach no broader support-package figure to that key, and on 3 November we removed the CPS row's proposed key (see above). The lane ran no git state-changing commands, commits or pushes. diff --git a/sources/harvest-uk-ab2026-2026-10-29/uk_tpa/NOTES.md b/sources/harvest-uk-ab2026-2026-10-29/uk_tpa/NOTES.md index 8bbbef53d..af2615c94 100644 --- a/sources/harvest-uk-ab2026-2026-10-29/uk_tpa/NOTES.md +++ b/sources/harvest-uk-ab2026-2026-10-29/uk_tpa/NOTES.md @@ -1,26 +1,26 @@ # uk_tpa: Tax Policy Associates, Autumn Budget 2026 -Dan Neidle's "Budget 2026 – what it means" (taxpolicy.org.uk, 28 October 2026), the seven chart data files the article embeds, and the calculator screenshot its footnote 1 links. Source slug `tax_policy_associates`, as in the AB2025 `uk_tpa` family. +In this family we stage Dan Neidle's "Budget 2026 – what it means" (taxpolicy.org.uk, 28 October 2026), the seven chart data files the article embeds, and the calculator screenshot its footnote 1 links. We use source slug `tax_policy_associates`, as in the AB2025 `uk_tpa` family. -**Working files.** `stage.py`, the file meant wherever these notes say `stage.py`, lives on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/thinktanks-a/uk_tpa/stage.py`; this branch carries only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz`. The lane's local primaries folder (`.ab2026-primaries/uk_tpa/`) was never committed: the repository carries no primary bytes, so fetch each primary from its `manifest.jsonl` URL and check its sha256. `stage.py` reads `uk_hmt_tables` from its sibling directory, so run it from a checkout where the family sits beside that one. +We keep `stage.py` on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/thinktanks-a/uk_tpa/stage.py`. We commit only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz` to this branch, and no primary bytes to the repository. The lane never committed its local primaries folder (`.ab2026-primaries/uk_tpa/`), so fetch each primary from its `manifest.jsonl` URL and check its sha256. `stage.py` reads `uk_hmt_tables` from its sibling directory, so run it from a checkout where the two families share a parent. -`stage.py` holds the rows as data and rewrites `manifest.jsonl` and `claims_staged.jsonl.gz` byte for byte. `stage.py --verify` checks the primaries' hashes, finds every quote in its primary, reads every chart value back from its JSON path, and re-derives each TPA group's Table 6.1 lines from `../uk_hmt_tables` (result: 0 problems). +We wrote the rows into `stage.py` as data, and it rewrites `manifest.jsonl` and `claims_staged.jsonl.gz` byte for byte. `stage.py --verify` checks the primaries' hashes and finds every quote in its primary. It also reads every chart value back from its JSON path and re-derives each TPA group's Table 6.1 lines from `../uk_hmt_tables`; it reports 0 problems. -- **Rows:** 305 (own 244, restated 61). -- **Primaries:** 10 (the article, seven chart data files, one PNG, and the calculator page as a 0-row descriptor). +- Rows: 305 (own 244, restated 61). +- Primaries: 10 (the article, seven chart data files, one PNG, and the calculator page as a 0-row descriptor). ## Access recipe -Every primary was fetched directly on 2026-10-29 between 16:17Z and 16:19Z, with `curl -sSL` and the browser User-Agent in the brief (HTTP 200). The bytes are under `.ab2026-primaries/uk_tpa/`; nothing came from a local copy. +We fetched every primary directly on 2026-10-29 between 16:17Z and 16:19Z, with `curl -sSL` and the browser User-Agent in the brief (HTTP 200). The lane saved the bytes under `.ab2026-primaries/uk_tpa/`, and we took nothing from a local copy. -- **The article** carries `article:published_time` 2026-10-28T15:51Z and `article:modified_time` 2026-10-29T09:59Z, so TPA edited it once after publication. The copy hashed here is the 16:17Z fetch. -- **The article changed again on 4 November.** Review B's fetch that day (217,180 bytes, sha256 `d95a0e8f8a0e6465…`) carries `article:modified_time` 2026-11-04T10:10:31Z. Five reader comments account for 15,211 of the 15,247 added bytes; all 49 quotes and footnotes 1–15 are unchanged, and it holds no figure the rows and these notes don't know. The article's own text and figures didn't change, so it is not a second primary under README rule 7 (the coordinator's call), and the rows keep citing the 29 October bytes. Like every HTML page, the article embeds per-request content, so a later fetch's sha256 differs; the check is that the quotes are still there. -- **The charts** are ECharts embeds. Each chart's `data-json` attribute names a JSON file under `/wp-content/assets/`, and those files are the primaries, URLs as named (three carry a `?v=20261028-n` cache-buster). As in AB2025, `doc_type` is `flourish_json`, the nearest enum for a chart-data file. -- **Values come from the data files, not from bar heights.** Where the chart prints a data label (`label.formatter`), the label is kept in `conditions.chart_label`. -- **Quotes** from a JSON file are the element re-serialised with `json.dumps`, with `£` decoded to £ (the AB2025 `uk_tpa` precedent); the file itself may print the same element across several lines. Three of the chart files print £ as the JSON unicode escape `\u00a3`. -- **Chart descriptions:** only four of the seven chart files carry an `aria.label.description`; the net totals are staged from those that do. -- **The PNG** (`child_benefit_budget_2026.png`, linked from footnote 1) is a calculator screenshot. Its values are printed text in the tooltip panel, read off the image (`parse_confidence: medium`). -- **The "Download as PDF" button** builds a PDF in the browser (`tpa-pdf-exporter-loader.js`). It is not a separate document. +- TPA stamps the article with `article:published_time` 2026-10-28T15:51Z and `article:modified_time` 2026-10-29T09:59Z, so it edited the article once after publication. We hashed the 16:17Z fetch. +- By 4 November readers had added five comments to the article: review B fetched the article that day (217,180 bytes, sha256 `d95a0e8f8a0e6465…`) and found `article:modified_time` 2026-11-04T10:10:31Z. The comments account for 15,211 of the 15,247 added bytes. In review B's 4 November fetch the 49 quotes and footnotes 1–15 read as on 29 October, and TPA had added no figure that the rows and these notes don't already cover. Because TPA changed neither the article's own text nor its figures, the coordinator ruled it not a second primary under README rule 7, and we keep citing the 29 October bytes in the rows. Like every HTML page in this harvest, the article embeds per-request content, so a later fetch's sha256 differs; check instead that the quotes still appear. +- TPA draws the charts as ECharts embeds. In each chart's `data-json` attribute TPA names a JSON file under `/wp-content/assets/`, and we take those files as the primaries, at the URLs TPA names (three end in a `?v=20261028-n` cache-buster). As in AB2025, we set `doc_type` to `flourish_json`, the nearest enum for a chart-data file. +- We take values from the data files, not from bar heights. Where TPA prints a data label on the chart (`label.formatter`), we keep the label in `conditions.chart_label`. +- To quote from a JSON file, we re-serialise the element with `json.dumps` and decode `£` to £ (the AB2025 `uk_tpa` precedent); the file itself may print the same element across several lines. In three of the chart files TPA writes £ as the JSON unicode escape `\u00a3`. +- TPA gives only four of the seven chart files an `aria.label.description`, and we stage the net totals from those four. +- TPA links the PNG (`child_benefit_budget_2026.png`) from footnote 1, as a calculator screenshot. TPA prints its values as text in the tooltip panel, and we read them off the image (`parse_confidence: medium`). +- TPA's "Download as PDF" button builds a PDF in the browser (`tpa-pdf-exporter-loader.js`), so TPA publishes no separate document there. ## Rows by primary @@ -41,8 +41,8 @@ Every primary was fetched directly on 2026-10-29 between 16:17Z and 16:19Z, with | measure_key | rows | |---|--:| -| `null` (65 with a proposed key, 139 context) | 204 | | `ab2026__package_total_tax_policy_decisions` | 40 | +| the 14 group keys below | 65 | | `ab2026__writing_down_allowance_12pc_and_fya_50pc` | 15 | | `ab2026__bank_surcharge_6pc_temporary` | 11 | | `ab2026__hmrc_tax_gap_package` | 8 | @@ -51,10 +51,11 @@ Every primary was fetched directly on 2026-10-29 between 16:17Z and 16:19Z, with | `ab2026__foreign_sovereign_investments_taxation` | 4 | | `ab2026__electricity_vat_zero_rate_to_july_2027` | 2 | | 14 other single measures (Table 6.1 lines 3, 5, 15, 26, 44, 46, 49, 56, 59, 60, 61, 62, 86, 96), one row each | 14 | +| `null` (context) | 139 | -The 139 context rows have no measure key and no proposal: +We gave the 139 context rows no measure key: -- the earlier fiscal events in the two history charts (`conditions.policy_event`). The 2026 Budget series' 0.0 fills for 2024-25 and 2025-26, years Table 6.1 has no column for, carry the series' key (rows 98, 99 and 130; review B, TPA finding 1); +- the earlier fiscal events in the two history charts (`conditions.policy_event`). We key the 2026 Budget series' 0.0 fills for 2024-25 and 2025-26, years Table 6.1 has no column for, to the series' key instead (rows 98, 99 and 130; review B, TPA finding 1); - the other countries, the pre-Budget UK, and the after-Budget UK's bank levy and VAT components (which the surcharge leaves unchanged) in the bank chart; - the fiscal-context levels in footnote 2; - capital-allowance present values under pre-Budget regimes; @@ -62,155 +63,155 @@ The 139 context rows have no measure key and no proposal: ## Attribution: how TPA's numbers divide -TPA builds no yield estimates of its own. Every £ yield in the article and its charts is HM Treasury's Table 6.1 costing (or the OBR's), and TPA uses it in one of two ways: +TPA builds no yield estimates of its own. It takes every £ yield in the article and its charts from HM Treasury's Table 6.1 costing (or the OBR's), and uses it in one of two ways: -- **Repeats it,** for example a chart group that is exactly one Table 6.1 line, or "about £0.7bn" for the £695m total. These rows are `restated`, with `source_model: hmt_costing_obr_certified` (or `obr_efo_forecast` for the OBR's paragraph 3.17 averages) and `conditions.originator`. Every restated row's `note` names whose figure it is ("Restates …"), and says "TPA names no source" where TPA cites none. -- **Re-expresses it.** TPA groups lines, sums them over years, deflates them to 2026-27 prices, or divides them by GDP. These rows are `own`, with `source_model: arithmetic` and a `conditions.derived_from` naming the HMT or OBR input, and every note says "arithmetic on HM Treasury's costings, not an independent estimate". This matches the IFS lane's treatment of IFS's own aggregate of HMT costings. The ingest can separate these rows on `derived_from` if it treats them as restatements. +- **TPA repeats it**, for example when it gives a single Table 6.1 line its own group in a chart or calls the £695m total "about £0.7bn". We mark these rows `restated`, with `source_model: hmt_costing_obr_certified` (or `obr_efo_forecast` for the OBR's paragraph 3.17 averages) and `conditions.originator`. In every restated row's `note` we name whose figure it is ("Restates …"), and write "TPA names no source" where TPA cites none. +- **TPA re-expresses it**: it groups lines, sums them over years, deflates them to 2026-27 prices, or divides them by GDP. We mark these rows `own`, with `source_model: arithmetic`, name the HMT or OBR input in `conditions.derived_from`, and in every note say "arithmetic on HM Treasury's costings, not an independent estimate". We treat IFS's own aggregate of HMT costings the same way in the IFS family. At ingest we score all 186 of these `own` rows as TPA's claims, not restatements, or drop them on other grounds (see "Review fixes, 3 November 2026", finding 2). -TPA's own calculations from statutory parameters are `own` with no `derived_from`: +We mark TPA's own calculations from statutory parameters `own`, with no `derived_from`: - capital-allowance present values; - the Pension Tax Allowance's 53% marginal rate; - the notional investment bank's tax rates; - the calculator outputs. -Under README rule 8, a measure's own parameters are its design, not rows. The allowance's "£282 withdrawal band", "up to £94 a year" maximum and the "£93" it removes therefore sit in the conditions of the 53% row. +Under README rule 8 we treat a measure's own parameters as its design and stage no rows for them. We record the allowance's "£282 withdrawal band", "up to £94 a year" maximum and the "£93" it removes in the conditions of the 53% row. ## Conventions -- **Sign:** `positive = yield to the Exchequer` on every revenue row. The charts draw tax rises red and cuts green, which matches. -- **Units:** - - £ figures are in pounds (`£ billion × 1e9`). - - Charted % of GDP figures are percentages with `proposed_unit: percent_of_gdp`, the OBR lane's spelling. `conditions.denominator` says "nominal GDP". - - Present values are shares of cost. - - Marginal and effective rates are percent. -- **Period and price basis:** fiscal-year rows carry `fy`, with `period` the end year. Price bases are not converted: - - the history chart is in "2026 money"; - - the by-measure chart and footnote 7 are in "2026-27 prices"; - - every other £ figure is cash. -- **Cumulative rows:** the six-year totals sit at `fy` 2031-32 (`value_kind: cumulative`, `horizon` "2026-27 to 2031-32 (cumulative)"). -- **Present values and the bank chart:** the capital-allowance present values are properties of a regime, so they are `point_in_time` 2027, with the regime in `conditions.regime`. The bank chart is under "2027 rules" (`annual` 2027). -- **Rows with no stated year:** "about £1bn a year in the medium term", HM Treasury's 84% (row 9, staged at 2027-28 as uk_hmt_redbook stages para 2.50) and similar figures sit at a stated year, with a `year_note` and `parse_confidence: medium`. The OBR's paragraph 3.17 averages that footnote 8 quotes (rows 29–30, "a year on average from 2028-29") sit at 2031-32, the forecast's final year, as the uk_obr_tables family indexes the same sentences. Footnote 15's "£450m to £495m a year" (a range across 2028-29 to 2031-32) is staged at the years Table 6.1 line 68 shows those values: £450m in 2029-30 and £495m in 2028-29. -- **Value kinds:** TPA states no ranges for its own estimates, so "About two-fifths" is a `point`, like the family's other "about" figures. -- **Geography:** the calculator rows use AB2025's code `UK_excl_scotland`; the bank chart's other countries are named. -- **Parts (README rule 9):** the bank chart's "Corporate income taxes" for the after-Budget UK (row 278) is a printed part of the 33.7% total (row 268), so it is a `measure_component`. The OBR's £0.4bn (row 30) is Table 6.1 line 61's receipts head, so it is a `measure_component` of that line ("Tax head (receipts), annual average 2028-29 to 2031-32 (para 3.17)"). +- **Sign.** We set `positive = yield to the Exchequer` on every revenue row. TPA draws tax rises red and cuts green, which matches. +- **Units.** + - We give £ figures in pounds (`£ billion × 1e9`). + - We give charted % of GDP figures as percentages with `proposed_unit: percent_of_gdp`, the OBR lane's spelling, and set `conditions.denominator` to "nominal GDP". + - We give present values as shares of cost. + - We give marginal and effective rates in percent. +- **Period and price basis.** We give fiscal-year rows `fy`, with `period` the end year. We convert no price bases: + - TPA gives the history chart in "2026 money"; + - TPA gives the by-measure chart and footnote 7 in "2026-27 prices"; + - TPA gives every other £ figure in cash. +- **Cumulative rows.** We put the six-year totals at `fy` 2031-32 (`value_kind: cumulative`, `horizon` "2026-27 to 2031-32 (cumulative)"). +- **Present values and the bank chart.** In the capital-allowance present values TPA describes a regime, so we stage them `point_in_time` 2027, with the regime in `conditions.regime`. TPA draws the bank chart under "2027 rules" (`annual` 2027). +- **Rows with no stated year.** We put "about £1bn a year in the medium term", HM Treasury's 84% (row 9, staged at 2027-28 as we stage para 2.50 in uk_hmt_redbook) and similar figures at a stated year, with a `year_note` and `parse_confidence: medium`. We put the OBR's paragraph 3.17 averages that TPA quotes in footnote 8 (rows 29–30, "a year on average from 2028-29") at 2031-32, the forecast's final year, as we index the same sentences in the uk_obr_tables family. In footnote 15 TPA gives "£450m to £495m a year" as a range across 2028-29 to 2031-32. We stage it at the years in which HM Treasury shows those values on Table 6.1 line 68: £450m in 2029-30 and £495m in 2028-29. +- **Value kinds.** TPA states no ranges for its own estimates, so we stage "About two-fifths" as a `point`, like the family's other "about" figures. +- **Geography.** We give the calculator rows AB2025's code `UK_excl_scotland`, and name the bank chart's other countries. +- **Parts (README rule 9).** TPA prints the bank chart's "Corporate income taxes" for the after-Budget UK (row 278) as a part of the 33.7% total (row 268), so we mark it a `measure_component`. The OBR's £0.4bn (row 30) covers only Table 6.1 line 61's receipts head, so we mark it a `measure_component` of that line ("Tax head (receipts), annual average 2028-29 to 2031-32 (para 3.17)"). -## How each TPA group's Table 6.1 lines were found +## How we found each TPA group's Table 6.1 lines -TPA's footnote 6 says "Our mapping of each row to a group is available on request", so the mapping isn't published. The lines below were inferred by matching TPA's printed values to Table 6.1 (`../uk_hmt_tables`). `stage.py --verify` repeats the match: +In footnote 6 TPA says "Our mapping of each row to a group is available on request", so TPA hasn't published the mapping. We inferred the lines below by matching TPA's printed values to Table 6.1 (`../uk_hmt_tables`), and `stage.py --verify` repeats the match: -- **By-year chart:** every group reproduces TPA's value in all six years exactly. Its 11 groups partition the 53 Table 6.1 tax-head lines, which sum to £700m in 2027-28, as footnote 5 says. -- **By-tax chart (2027-28 only):** each group matches its 2027-28 value exactly. Lines that are zero in 2027-28 are assigned by the by-year mapping and, for line 95, by the by-measure "Tariffs" bar. One line stays undetermined: line 66 (the summer 2026 cut to 5% VAT on family days out). It is zero in 2027-28, the by-year chart puts it in "Temporary VAT and fuel cuts", and the by-tax chart splits that group into electricity VAT and fuel, red diesel and HGV VED, neither of which it fits. It may sit in the by-tax "Other VAT" or "Other", and it is left out of both scopes below. The values are unaffected. -- **By-measure chart (real terms):** deflators fitted to its 19 single-measure bars (1.000, 0.9705, 0.9503, 0.9310, 0.9133, 0.8943 for 2026-27 to 2031-32) reproduce every bar to within £1m. That settles the multi-line bars: "Savings and property income tax rates +2ppts" is lines 61 and 62 (3.981; adding line 11 would give 3.816), "Tariffs" includes line 95 (−1.594; without it −1.776), and "Other" is the remainder. +- **By-year chart.** Every group reproduces TPA's value in all six years exactly. TPA's 11 groups partition the 53 Table 6.1 tax-head lines, which sum to £700m in 2027-28, as TPA says in footnote 5. +- **By-tax chart (2027-28 only).** Each group matches its 2027-28 value exactly. We assign the lines that are zero in 2027-28 from the by-year mapping and, for line 95, from the by-measure "Tariffs" bar. We couldn't determine one line: line 66 (the summer 2026 cut to 5% VAT on family days out). HM Treasury shows zero on it in 2027-28, and in its by-year chart TPA puts it in "Temporary VAT and fuel cuts". But in its by-tax chart TPA splits that group into electricity VAT and fuel, red diesel and HGV VED, and line 66 fits neither. TPA may count it in the by-tax "Other VAT" or "Other", so we leave it out of both scopes below. Leaving line 66 out changes no value. +- **By-measure chart (real terms).** We fitted deflators to its 19 single-measure bars (1.000, 0.9705, 0.9503, 0.9310, 0.9133, 0.8943 for 2026-27 to 2031-32), and they reproduce every bar to within £1m. With them we read the multi-line bars. TPA puts lines 61 and 62 in "Savings and property income tax rates +2ppts" (3.981; adding line 11 would give 3.816), line 95 in "Tariffs" (−1.594; without it −1.776), and the remainder in "Other". -A chart group that is exactly one Table 6.1 line carries that line's registry key. Other groups carry a proposed key. +When TPA gives a single Table 6.1 line its own group in a chart, we give the group that line's registry key. On 29 October we gave the other groups `measure_key: null` and proposed the 14 keys below, and in tranche 2 we registered them. -## Proposed measure keys (all `measure_key: null`) +## Group keys -| proposed key | Table 6.1 lines | TPA's words | rows | +| key | Table 6.1 lines | TPA's words | rows | |---|---|---|--:| | `ab2026__package_property_and_savings_income_rates` | 61, 62 | "The savings and property rise"; "Savings and property income tax rates +2ppts" (by-measure chart). Same spelling and scope as the uk_hmt_redbook family | 5 | -| `ab2026__package_tpa_savings_property_income_and_cash_isa` | 11, 61, 62 | "Savings and property income" (by-year), "Savings and property income tax" (by-tax). The group includes line 11, the cash ISA limit held at £20,000, which TPA doesn't mention | 7 | +| `ab2026__package_tpa_savings_property_income_and_cash_isa` | 11, 61, 62 | "Savings and property income" (by-year), "Savings and property income tax" (by-tax). TPA puts line 11 ("Maintain the cash ISA limit at £20,000") in the group without mentioning it | 7 | | `ab2026__package_tpa_temporary_vat_and_fuel_cuts` | 2, 3, 18, 65, 66 | "Temporary VAT and fuel cuts" (by-year) | 6 | | `ab2026__package_tpa_fuel_duty_red_diesel_and_hgv_ved` | 3, 18, 65 | "Fuel duty cut, red diesel and HGV VED (temporary)" (by-tax) | 1 | -| `ab2026__package_tpa_business_rates` | 12, 13, 14, 57, 76, 80 | "Business rates" (all three charts). Registry caveat (review B, TPA finding 5): on the by-tax bar, line 76, zero in 2027-28, is placed by assumption; the values do not depend on it. | 8 | +| `ab2026__package_tpa_business_rates` | 12, 13, 14, 57, 76, 80 | "Business rates" (all three charts). Registry caveat (review B, TPA finding 5): on the by-tax bar, we place line 76, zero in 2027-28, by assumption; the values do not depend on it. | 8 | | `ab2026__package_tpa_energy_and_environment` | 17, 69, 71, 90, 96 | "Energy and environment" (by-year) | 6 | -| `ab2026__package_tpa_other_income_tax` | 5, 15, 28, 52, 53, 86, 94 | "Other income tax" (by-year); "Income tax: mileage rates, Pension Tax Allowance and other" (by-tax). Registry caveat (review B, TPA finding 5): on the by-tax bar, lines 15 and 86, zero in 2027-28, are placed by assumption; the values do not depend on them. | 7 | -| `ab2026__package_tpa_other_vat` | 36, 46, 51, 59, 88, 89 (the by-tax bar may also hold line 66, zero in 2027-28) | "Other VAT" (by-year, by-tax). Registry caveat (thinktanks-a review, finding 6): the by-tax lines that are zero in 2027-28 are placed by assumption and line 66 is undetermined; the values do not depend on it. | 7 | -| `ab2026__package_tpa_other_corporate_tax` | 49, 50, 54, 68, 97 | "Other corporate tax" (by-year, by-tax). Registry caveat (review B, TPA finding 5): on the by-tax bar, lines 49 and 50, zero in 2027-28, are placed by assumption; the values do not depend on them. | 7 | +| `ab2026__package_tpa_other_income_tax` | 5, 15, 28, 52, 53, 86, 94 | "Other income tax" (by-year); "Income tax: mileage rates, Pension Tax Allowance and other" (by-tax). Registry caveat (review B, TPA finding 5): on the by-tax bar, we place lines 15 and 86, zero in 2027-28, by assumption; the values do not depend on them. | 7 | +| `ab2026__package_tpa_other_vat` | 36, 46, 51, 59, 88, 89 (TPA may also put line 66, zero in 2027-28, in the by-tax bar) | "Other VAT" (by-year, by-tax). Registry caveat (thinktanks-a review, finding 6): we place the by-tax lines that are zero in 2027-28 by assumption, and couldn't determine line 66; the values do not depend on it. | 7 | +| `ab2026__package_tpa_other_corporate_tax` | 49, 50, 54, 68, 97 | "Other corporate tax" (by-year, by-tax). Registry caveat (review B, TPA finding 5): on the by-tax bar, we place lines 49 and 50, zero in 2027-28, by assumption; the values do not depend on them. | 7 | | `ab2026__package_tpa_other_by_year` | 10, 16, 19, 26, 44, 56, 60, 77, 78, 79, 93, 95, 98 | "Other" (by-year) | 6 | -| `ab2026__package_tpa_tariffs` | 10, 16, 77, 78, 95 | "Tariffs (incl. UK-India)" (by-tax, by-measure). Registry caveat (review B, TPA finding 5): on the by-tax bar, line 95, zero in 2027-28, is placed by assumption (the by-measure bar settles it); the values do not depend on it. | 2 | +| `ab2026__package_tpa_tariffs` | 10, 16, 77, 78, 95 | "Tariffs (incl. UK-India)" (by-tax, by-measure). Registry caveat (review B, TPA finding 5): on the by-tax bar, we place line 95, zero in 2027-28, by assumption (the by-measure bar settles it); the values do not depend on it. | 2 | | `ab2026__package_tpa_cgt_founders_relief_and_badr` | 26, 60 | "CGT (Founders' Relief, BADR)" (by-tax) | 1 | -| `ab2026__package_tpa_other_by_tax_2027_28` | 17, 19, 44, 56, 69, 71, 79, 90, 93, 96, 98 (line 66 undetermined) | "Other" (by-tax). The non-zero 2027-28 lines are 19, 44, 56, 69, 79, 90 and 93; the zero lines come from the by-year mapping. Registry caveat (thinktanks-a review, finding 6): the by-tax lines that are zero in 2027-28 are placed by assumption and line 66 is undetermined; the values do not depend on it. | 1 | +| `ab2026__package_tpa_other_by_tax_2027_28` | 17, 19, 44, 56, 69, 71, 79, 90, 93, 96, 98 (line 66 undetermined) | "Other" (by-tax). The non-zero 2027-28 lines are 19, 44, 56, 69, 79, 90 and 93; we take the zero lines from the by-year mapping. Registry caveat (thinktanks-a review, finding 6): we place the by-tax lines that are zero in 2027-28 by assumption, and couldn't determine line 66; the values do not depend on it. | 1 | | `ab2026__package_tpa_other_by_measure` | 11, 17, 18, 19, 28, 36, 50, 51, 53, 65, 66, 69, 71, 79, 88, 89, 90, 93, 94, 97, 98 | "Other" (by-measure) | 1 | -`ab2026__package_total_tax_policy_decisions` is registered. It carries TPA's rows for "today's Budget" and "the tax measures", the net totals, and the 2026 Budget series of the history charts, because their scope is all Table 6.1 tax lines. The series' 0.0 cells for 2024-25 and 2025-26 (rows 98, 99 and 130) carry the key too; Table 6.1 has no column for those years. The totals come in two kinds: +We use the registered `ab2026__package_total_tax_policy_decisions` for TPA's rows for "today's Budget" and "the tax measures", the net totals, and the 2026 Budget series of the history charts, because TPA scopes them to all Table 6.1 tax lines. We also key the series' 0.0 cells for 2024-25 and 2025-26 (rows 98, 99 and 130) to it, though Table 6.1 has no column for those years. TPA gives two kinds of total: - **Sums of the 53 tax lines** (`own`, `arithmetic`): footnote 5's £700m, the by-tax chart's "Net £0.7bn", footnote 7's £19.1bn and £21.0bn, and the by-measure chart's net. These differ from HM Treasury's printed totals by rounding only (£700m against £695m in 2027-28). -- **HM Treasury's printed totals** (`restated`): the by-year chart's "Net total" (it prints £-1.65bn for 2026-27, HMT's −£1,645m; the 53 lines sum to −£1,635m), the GDP chart's £ figures, footnote 5's £695m and the article's "about £0.7bn". +- **HM Treasury's printed totals** (`restated`): the by-year chart's "Net total" (TPA prints £-1.65bn for 2026-27, HMT's −£1,645m; the 53 lines sum to −£1,635m), the GDP chart's £ figures, footnote 5's £695m and the article's "about £0.7bn". ## Proposed metrics and units -Reused from the AB2025 README list: `effective_tax_rate` (the notional bank's tax as a % of profit). +We reuse one metric from the AB2025 README list: `effective_tax_rate` (the notional bank's tax as a % of profit). -New proposals, each with TPA's own words: +We propose these new metrics, each with TPA's own words: - **`capital_allowance_present_value` (share):** "the present value of the allowances as a percentage of the cost, discounting at 5% a year" (footnote 12). - **`marginal_tax_rate` (percent):** "a 53% marginal rate for pensioners on about £13,000", and the calculator's marginal rate. - **`marginal_take_home_pay` (gbp):** "Of the next £1,000, you keep £417" (calculator screenshot; a level at the margin under current rules, not a policy change). - **`marginal_tax_amount` (gbp):** "Income tax £400", "NI £20" and "Child benefit charge £163" on the next £1,000 of salary (same screenshot). -- **`unaffected_share` (share):** "84% of income tax payers will be unaffected". This is not `affected_share`: the value is the unaffected share. +- **`unaffected_share` (share):** "84% of income tax payers will be unaffected". We don't use `affected_share`, because TPA gives the unaffected share. - **`tax_to_gdp_ratio` (`percent_of_gdp`):** "tax burden … 33% of GDP", "Tax take". - **Context levels from footnote 2:** `public_sector_net_debt_change` (percentage points), `programme_cost` (gbp) and `debt_interest_cost` (gbp, the uk_thinktank_misc spelling). -- **`revenue_share_by_fiscal_event` (share):** "About two-fifths of that comes from Conservative Budgets", the share of the £123bn of extra tax in 2030-31 coming from a group of Budgets. The registered `revenue_share` is a group's share of one measure's yield, so it does not fit. +- **`revenue_share_by_fiscal_event` (share):** "About two-fifths of that comes from Conservative Budgets", the share of the £123bn of extra tax in 2030-31 coming from a group of Budgets. We register `revenue_share` for a group's share of one measure's yield, so it doesn't fit here. -The family uses the registered Metrics `revenue_change`, `caseload` and, for the totals of the Budget's tax measures, `decisions_effect_on_borrowing` (README rule 10, below). No row proposes a baseline (`baseline_policy` is null throughout). +We use the registered metrics `revenue_change`, `caseload` and, for the totals of the Budget's tax measures, `decisions_effect_on_borrowing` (README rule 10, below). We propose no baseline on any row (`baseline_policy` is null throughout). ## Not staged, and why -- **Policy parameters** (README rule 8, "stage estimates, not design"): +- We left out policy parameters (README rule 8, "stage estimates, not design"): - the Pension Tax Allowance's "up to £94 a year" maximum and "£282 withdrawal band" (in the conditions of the 53% row); - - "£93", the most tax the allowance removes (20% × (£13,036.40 − £12,570) = £93.28), the same quantity as the £94 maximum: staged on 29 October, dropped on 3 November and kept as `conditions.tax_removed` on the 53% row; + - "£93", the most tax the allowance removes (20% × (£13,036.40 − £12,570) = £93.28), the same quantity as the £94 maximum. We staged it on 29 October, dropped it on 3 November and kept it as `conditions.tax_removed` on the 53% row; - savings and property rates of 24%, 44% and 49%; - writing-down allowance rates of 18%, 14%, 12% and 6%, and first-year allowances of 40% and 50%; - the £1m annual investment allowance; - - the full new State Pension of £13,036.40 (carried in a condition) and the £1-for-£3 taper; + - the full new State Pension of £13,036.40 (which we record in a condition) and the £1-for-£3 taper; - the 15% REIT rate in footnote 14; - the £60-80k child benefit band in footnote 1; - - the 2.0% CPI assumption for 2032 (carried in `conditions.deflator`). -- **Duplicate phrasings** in one primary: "a new 53% band, for £94" repeats the paragraph's own 53% and £94. -- **Figures without a year or a measure:** + - the 2.0% CPI assumption for 2032 (which we record in `conditions.deflator`). +- We left out duplicate phrasings in one primary: TPA's "a new 53% band, for £94" repeats the paragraph's own 53% and £94. +- We left out figures without a year or a measure: - "the UK's approximately £3,000bn GDP": an approximate OBR level with no year. - "over twice its pre-pandemic share of GDP": a ratio with no printed base. -- **Not numeric:** the OBR's "High" uncertainty rating for the property costing (footnote 8) is categorical. It belongs to the OBR family. -- **Out of scope:** the calculator page's 69.5% marginal rate under Reform UK's £15,000 allowance is an opposition proposal, not the Budget's. -- **Not TPA's:** - - image-6.png reproduces Red Book paragraph 2.54, HM Treasury's text, which holds only parameters. - - Reader comments under the article, for example "the cost is £1.2bn" for Founders' Relief, are not TPA's. -- **Further-reading thumbnails** (image-2, investment_bank_tax_as_of_profit-1, taxable_capital_gains_…) illustrate other articles. -- **The calculator page** (25 October, updated for the Budget) is a 0-row descriptor, as in AB2025: - - it calculates take-home pay, marginal and effective rates, and net-v-gross for employment and other income, current rules against Budget 2027/28; - - its worked examples (62% in the £100k taper, 67.3% and 73.3% with three children and student loans, 84.5% and 89.3% in Scotland, 2,000,000% at the childcare cliff) describe the tax system, not the Budget's measures; - - it restates CenTax's £640m childcare costing, which is staged in uk_centax. - - Its on-screen outputs are computed in the browser and are not printed text. -- **Pre-Budget TPA articles linked from this one, outside the assignment:** "Banks say they're overtaxed. The TUC says they're undertaxed" (22 October, linked from the bank-tax paragraph), "Capital gains tax reform was right. Now it's dead." (6 October), the Cebr VAT-free shopping article (28 September) and "37 ways Andy Burnham could raise £4.7bn" (3 July). They are candidates for a follow-up if the family is to include pre-Budget options, as AB2025's did. +- We left out the OBR's "High" uncertainty rating for the property costing, which TPA quotes in footnote 8; the OBR rates the costing's uncertainty by category, so the rating belongs in the OBR family. +- We left out the calculator page's 69.5% marginal rate under Reform UK's £15,000 allowance, which falls out of scope as an opposition proposal, not the Budget's. +- We left out material that isn't TPA's: + - In image-6.png TPA reproduces Red Book paragraph 2.54, HM Treasury's text, in which it gives only parameters. + - Readers, not TPA, wrote the comments under the article, for example "the cost is £1.2bn" for Founders' Relief. +- We left out the further-reading thumbnails (image-2, investment_bank_tax_as_of_profit-1, taxable_capital_gains_…), which illustrate other articles. +- We manifest the calculator page (25 October, updated for the Budget) as a 0-row descriptor, as in AB2025: + - TPA's calculator works out take-home pay, marginal and effective rates, and net-v-gross for employment and other income, current rules against Budget 2027/28; + - TPA's worked examples on it (62% in the £100k taper, 67.3% and 73.3% with three children and student loans, 84.5% and 89.3% in Scotland, 2,000,000% at the childcare cliff) describe the tax system, not the Budget's measures; + - TPA restates CenTax's £640m childcare costing there, which we stage in uk_centax. + - The browser computes its on-screen outputs, and TPA prints none of them as text. +- We left out pre-Budget TPA articles that this one links, outside the assignment: "Banks say they're overtaxed. The TUC says they're undertaxed" (22 October, linked from the bank-tax paragraph), "Capital gains tax reform was right. Now it's dead." (6 October), the Cebr VAT-free shopping article (28 September) and "37 ways Andy Burnham could raise £4.7bn" (3 July). Each could seed a follow-up if the family should include pre-Budget options, as AB2025's did. ## Things in the primaries that look wrong or contradictory (reported, not fixed) - **Pension Tax Allowance maximum:** - - The article says "up to £94 a year" and, a paragraph later, "£93". - - OOTLAR paragraph 1.4, which TPA cites, gives the rule: 20% of the full new State Pension minus the Personal Allowance. On TPA's own £13,036.40, that is £93.28. - - The "£282 withdrawal band" is three times £94; three times £93.28 is £279.84. -- **Start year of the history total:** the £123bn sentence says tax changes "since 2021"; footnote 3 says measures "announced since March 2020". -- **OBR's property figure:** footnote 8 quotes the OBR (EFO paragraph 3.17) correctly, but the OBR's "£0.4 billion a year on average from 2028-29" for the property rise sits below Table 6.1 line 61's average over 2028-29 to 2031-32 (£478m). -- **Rounding:** the bank chart's stacked components sum within 0.1pp of the printed totals, for example the Netherlands at 43.7 against 43.8. + - TPA says "up to £94 a year" and, a paragraph later, "£93". + - In OOTLAR paragraph 1.4, which TPA cites, HM Treasury gives the rule: 20% of the full new State Pension minus the Personal Allowance. On TPA's own £13,036.40, that gives £93.28. + - TPA's "£282 withdrawal band" equals three times £94; three times £93.28 gives £279.84. +- **Start year of the history total:** in the £123bn sentence TPA says tax changes "since 2021"; in footnote 3 it says measures "announced since March 2020". +- **OBR's property figure:** in footnote 8 TPA quotes the OBR (EFO paragraph 3.17) correctly. But the OBR's "£0.4 billion a year on average from 2028-29" for the property rise falls short of Table 6.1 line 61's average over 2028-29 to 2031-32 (£478m). +- **Rounding:** the bank chart's stacked components sum to within 0.1pp of the printed totals, for example the Netherlands at 43.7 against 43.8. ## Review fixes, 3 November 2026 -The thinktanks-a review (`harvest-staging/ab2026/reviews/verify-thinktanks-a.md` on the staging branch) approved this family with fixes. Each sits in an explicit block at the end of `stage.py`'s rows. No primary was re-opened (the session's network policy refused taxpolicy.org.uk); no printed value changed. +In its review (`harvest-staging/ab2026/reviews/verify-thinktanks-a.md` on the staging branch), the thinktanks-a reviewer approved this family with fixes. We put each in an explicit block at the end of `stage.py`'s rows. We re-opened no primary, because the session's network policy refused taxpolicy.org.uk, and changed no printed value. -- **Finding 1 (the coordinator's call):** row 16, "£93", is the allowance's design under rule 8 and is dropped; row 17 (53%) carries `conditions.tax_removed: "£93 (as printed)"`. 306 rows become 305. -- **Finding 2:** no change. The group, deflated and %-of-GDP rows carry `conditions.derived_from`; tranche 2 separates them from model estimates on it. -- **Finding 3:** no change. The 2027-28 total tax policy decisions appear in several rows (3, 26, 27, 147, 169 and 237, 0-based before the drop) and the six-year total in rows 4, 28 and 265; tranche 2 should deduplicate them. -- **Finding 4:** the quote convention above now says JSON elements are re-serialised. -- **Finding 5 (optional, applied):** row 1 ("About two-fifths") uses `proposed_metric: revenue_share_by_fiscal_event`, not the registered `revenue_share`. -- **Finding 6:** the two by-tax scopes carry the registry caveat in the proposed-key table. -- **Verification:** the contract checker reports 305 rows, 0 problems, and a re-run of `stage.py` reproduces both files byte for byte. COUNTS.json pins the gzip's sha256. +- Finding 1, the coordinator's call: we dropped row 16, "£93", as the allowance's design under rule 8, and set `conditions.tax_removed: "£93 (as printed)"` on row 17 (53%). The 306 rows became 305. +- Finding 2: we made no change, since we had already set `conditions.derived_from` on the group, deflated and %-of-GDP rows. At ingest we copy the field onto each of the 155 claims these 186 rows become, and drop the other 31 (30 as out of scope, 1 as a ratio). No code reads the field, so separating these claims from model estimates stays open. +- Finding 3: we made no change. We stage the 2027-28 total tax policy decisions on rows 3, 26, 27, 147, 169 and 237 (0-based before the drop) and the six-year total on rows 4, 28 and 265. At ingest, counting today's 305 rows from 0, we drop the restated rows 3, 26, 146 and 236, and score rows 25 and 168 (2027-28) and 4, 27 and 264 (six years) on `decisions_effect_on_borrowing`. We merge none of these five at ingest, so their duplication stays open. +- Finding 4: we now say, in the quote convention above, that we re-serialise JSON elements. +- Finding 5 (optional, applied): on row 1 ("About two-fifths") we use `proposed_metric: revenue_share_by_fiscal_event`, not the registered `revenue_share`. +- Finding 6: we added the registry caveat to the two by-tax scopes in the group-key table. +- The contract checker reports 305 rows, 0 problems, and when we re-run `stage.py` it reproduces both files byte for byte. COUNTS.json pins the gzip's sha256. ## README rule 10, 3 November 2026 -Under README rule 10, `stage.py` moves TPA's figures for all the Budget's tax measures from `revenue_change` to `decisions_effect_on_borrowing`: the 37 rows keyed to `ab2026__package_total_tax_policy_decisions`, in pounds or as a share of GDP, and the three 0.0 cells that TPA's charts print for years before Table 6.1 starts, which now carry the key too (40 rows). Counting from 0, the rows are 2–4, 25–28, 98–104, 130–135, 143–154, 168, 235–240 and 264. Single measures and groups of measures keep `revenue_change`. No value or sign changed. +Under README rule 10, in `stage.py` we move TPA's figures for all the Budget's tax measures from `revenue_change` to `decisions_effect_on_borrowing`. We move the 37 rows keyed to `ab2026__package_total_tax_policy_decisions`, in pounds or as a share of GDP. We also move the three 0.0 cells that TPA prints in its charts for years before Table 6.1 starts, which we now key the same way (40 rows in all). Counting from 0, the rows are 2–4, 25–28, 98–104, 130–135, 143–154, 168, 235–240 and 264. We keep `revenue_change` for single measures and groups of measures, and changed no value or sign. ## Review fixes, 4 November 2026 -Two independent reviews of PR #231 at 4f424b9 approved this family with fixes, all low: review B (TPA findings 1–6) and review A (K-5, H-6(d) and H-6(e)). They sit in an explicit block after the rule 10 block in `stage.py`. No printed value changed. Indices are 4f424b9's. +Two independent reviews of PR #231 at 4f424b9 approved this family with fixes, all low: review B (TPA findings 1–6) and review A (K-5, H-6(d) and H-6(e)). We put them in an explicit block after the rule 10 block in `stage.py` and changed no printed value. We give 4f424b9's indices. -- **Finding 1 (the first fix, the coordinator's call):** rows 98, 99 and 130, the 2026 Budget series' 0.0 cells for 2024-25 and 2025-26, carry `ab2026__package_total_tax_policy_decisions`, with the note "Table 6.1 has no column for this year; TPA's chart prints 0.0". The key has 40 rows and the context rows go from 142 to 139. -- **Finding 2 (NOTES only, the coordinator's call):** the access recipe records the article's 4 November edit (reader comments only); it is not a second primary. -- **Finding 3 (and review A's H-6(e)):** the 21 restated rows 18–24, 26, 43–48 and 136–142 each name their originator in `note`, and rows 18, 20, 22 and 142 say "TPA names no source". Row 21's originator is the OBR's *Fiscal risks and sustainability*, July 2025, which footnote 2 links. -- **Finding 4:** row 278 (corporate income taxes, 30.5%) is a `measure_component` of row 268's 33.7%. -- **Finding 5:** the by-tax caveat is on the four further scopes (`…tpa_other_corporate_tax`, `…tpa_other_income_tax`, `…tpa_tariffs`, `…tpa_business_rates`), and the two existing caveats moved into the scope column. -- **Finding 6, nits:** rows 299–304 use `UK_excl_scotland` and record the tooltip's "Smoothed tapers" (`conditions.calculator_setting`); row 9 has a `year_note` and medium confidence; row 1 is a `point`; rows 10–12's locator names the third paragraph after the Red Book excerpt; the chart notes, the £ escapes and the Reform UK rate are recorded. Also applied: row 141 carries its printed `chart_label` (38.5%), which review B's sample noted. -- **Review A, K-5:** row 30 (the OBR's £0.4bn, line 61's receipts) is a `measure_component` with `component: "Tax head (receipts), annual average 2028-29 to 2031-32 (para 3.17)"`. The optional alignment is applied: rows 29 and 30 sit at 2031-32, as the OBR family's paragraph 3.17 rows do. -- **Verification:** the contract checker reports 305 rows, 0 problems, and a re-run of `stage.py` in a scratch directory reproduces both files byte for byte. +- Finding 1, the first fix and the coordinator's call: we key rows 98, 99 and 130, the 2026 Budget series' 0.0 cells for 2024-25 and 2025-26, to `ab2026__package_total_tax_policy_decisions`. We wrote in their note "Table 6.1 has no column for this year; TPA's chart prints 0.0". The key now has 40 rows, and the context rows went from 142 to 139. +- Finding 2, NOTES only and the coordinator's call: in the access recipe we record the article's 4 November change (five reader comments), and don't treat it as a second primary. +- Finding 3, and review A's H-6(e): in each of the 21 restated rows 18–24, 26, 43–48 and 136–142 we name the originator in `note`, and on rows 18, 20, 22 and 142 we say "TPA names no source". We name the OBR's *Fiscal risks and sustainability*, July 2025, which footnote 2 links, as row 21's originator. +- Finding 4: we mark row 278 (corporate income taxes, 30.5%) as a `measure_component` of row 268's 33.7%. +- Finding 5: we added the by-tax caveat to the four further scopes (`…tpa_other_corporate_tax`, `…tpa_other_income_tax`, `…tpa_tariffs`, `…tpa_business_rates`), and moved the two existing caveats into the scope column. +- Finding 6, nits: on rows 299–304 we use `UK_excl_scotland` and record the tooltip's "Smoothed tapers" (`conditions.calculator_setting`). We gave row 9 a `year_note` and medium confidence, and staged row 1 as a `point`. We name the third paragraph after the Red Book excerpt in rows 10–12's locator, and recorded the chart notes, the £ escapes and the Reform UK rate. We also gave row 141 its printed `chart_label` (38.5%), which review B noted in its sample. +- Review A, K-5: we mark row 30 (the OBR's £0.4bn, line 61's receipts) as a `measure_component` with `component: "Tax head (receipts), annual average 2028-29 to 2031-32 (para 3.17)"`. We also applied the optional alignment and put rows 29 and 30 at 2031-32, as we do for the OBR family's paragraph 3.17 rows. +- The contract checker reports 305 rows, 0 problems, and when we re-run `stage.py` in a scratch directory it reproduces both files byte for byte. diff --git a/sources/harvest-uk-ab2026-2026-10-29/uk_trussell_wpi/NOTES.md b/sources/harvest-uk-ab2026-2026-10-29/uk_trussell_wpi/NOTES.md index 0ed3687d2..6c836db82 100644 --- a/sources/harvest-uk-ab2026-2026-10-29/uk_trussell_wpi/NOTES.md +++ b/sources/harvest-uk-ab2026-2026-10-29/uk_trussell_wpi/NOTES.md @@ -1,12 +1,12 @@ # uk_trussell_wpi — Trussell, Autumn Budget 2026 -**Working files.** `stage.py`, which writes `manifest.jsonl` and `claims_staged.jsonl.gz`, lives on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/thinktanks-b/uk_trussell_wpi/stage.py`; this branch carries only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz`. The repository carries no primary bytes: fetch the primary from its `manifest.jsonl` URL and check its sha256. +We keep `stage.py`, which writes `manifest.jsonl` and `claims_staged.jsonl.gz`, on branch `uk/ab2026-harvest-staging` at `harvest-staging/ab2026/lanes/thinktanks-b/uk_trussell_wpi/stage.py`. We commit only NOTES.md, `manifest.jsonl` and `claims_staged.jsonl.gz` to this branch, and no primary bytes to the repository: fetch the primary from its `manifest.jsonl` URL and check its sha256. -Source `trussell_wpi`, as in AB2025's `uk_trussell_wpi`. **1 row from 1 primary.** The thinktanks-b lane staged this row in `uk_thinktank_misc` on 29 October with source `trussell`. On 3 November its review's finding 5 set the slug to AB2025's `trussell_wpi`, and the coordinator moved the row and its manifest line here, as AB2025 keeps Trussell's Budget responses in their own family. The row is otherwise unchanged. `stage.py` is `uk_thinktank_misc`'s script, filtered to Trussell's row and primary. +We use source `trussell_wpi`, as AB2025's `uk_trussell_wpi` does, and stage **1 row from 1 primary.** The thinktanks-b lane staged this row in `uk_thinktank_misc` on 29 October with source `trussell`. On 3 November, that family's review asked in finding 5 for AB2025's slug `trussell_wpi`, and the coordinator moved the row and its manifest line here, since AB2025 keeps Trussell's Budget responses in their own family. The coordinator changed nothing else in the row. We copied `uk_thinktank_misc`'s script into `stage.py` and filtered it to Trussell's row and primary. ## Access -The thinktanks-b lane found [Our response to the Autumn Budget 2026](https://www.trussell.org.uk/news-and-research/news/our-response-to-the-autumn-budget-2026) by searching Trussell's own news pages; the page prints “28th Oct 2026”. It was fetched fresh on 29 October 2026 with `curl -sSL` and the harvest's Safari User-Agent: HTTP 200, fetch completed 15:23:24Z, 171,343 bytes, sha256 `6cd2162d1382f03e937a382a39bd38c497a83d357bd969d2c85598740d17a95e`. The page's own body text was read; the hash is of the HTML. +The thinktanks-b lane found [Our response to the Autumn Budget 2026](https://www.trussell.org.uk/news-and-research/news/our-response-to-the-autumn-budget-2026) by searching Trussell's own news pages, and Trussell dates the page “28th Oct 2026”. The lane fetched it fresh on 29 October 2026 with `curl -sSL` and the harvest's Safari User-Agent: HTTP 200, fetch completed 15:23:24Z, 171,343 bytes, sha256 `6cd2162d1382f03e937a382a39bd38c497a83d357bd969d2c85598740d17a95e`. The lane read the page's own body text and hashed the HTML. ## Coverage and conventions @@ -14,17 +14,17 @@ The thinktanks-b lane found [Our response to the Autumn Budget 2026](https://www |---|---:|---| | `null` | 1 | Helen Barnard's “over half of private renters on low incomes” in receipt of Local Housing Allowance ran out of food and could not afford more, in the last 12 months | -- `metric: affected_share`, the registered metric (`models.py` registers it, so it goes in `metric`, not `proposed_metric`; changed on 4 November, review B finding 1 and review A's H-6(a)), unit `share`: “over half” is the strict lower bound 0.5, `value_kind: range_low`. -- Trussell calls the finding its own research but gives no method, research dates or exact percentage: `source_model: unstated`, `own`, `administrative_fact` for a retrospective reported observation, `parse_confidence: medium`. -- Context, not the modelled effect of the LHA freeze or of any Budget measure: no measure key and no proposed key. The 12-month window is `conditions.horizon`; the period is the publication-year anchor 2026, `point_in_time`. +- We record `metric: affected_share` with unit `share`. Because `models.py` registers that metric, we put it in `metric`, not `proposed_metric`; we made that change on 4 November, after review B finding 1 and review A's H-6(a). We read “over half” as the strict lower bound 0.5, `value_kind: range_low`. +- Trussell calls the finding its own research but gives no method, research dates or exact percentage. We therefore record `source_model: unstated`, `own`, `administrative_fact` for a retrospective reported observation, and `parse_confidence: medium`. +- Trussell reports context, not the modelled effect of the LHA freeze or of any Budget measure, so we gave the row no measure key and no proposed key. We put the 12-month window in `conditions.horizon` and use the publication-year anchor 2026, `point_in_time`, as the period. ## Not staged - Trussell prints no measured effect of the frozen LHA or of the Budget. -- Its statement that the government has already lifted the two-child limit is qualitative and has no row. +- Trussell states that the government has already lifted the two-child limit, a qualitative claim we staged no row for. ## Verification -- **4 November:** review B (PR #231, issue comment 6635498758) finding 1: the row carries the registered `affected_share` in `metric`, with `proposed_metric` null. The edit sits in the review-fixes block `stage.py` shares word for word with `uk_thinktank_misc`. The contract checker reports 1 row, 0 problems, and a re-run of `stage.py` reproduces both files byte for byte. No primary was re-opened. -- **3 November:** the contract checker reports 1 row, 0 problems; a re-run of `stage.py` reproduces `manifest.jsonl` and `claims_staged.jsonl.gz` byte for byte. No primary was re-opened (the session's network policy refused trussell.org.uk). -- **29 October (in `uk_thinktank_misc`):** the lane re-opened the saved page at the locator and checked the quote and value against it, and checked the quote is contiguous text under 15 words containing `value_raw`. +- On 4 November, review B (PR #231, issue comment 6635498758) raised finding 1, and we now record the registered `affected_share` in `metric`, with `proposed_metric` null. We made the edit in the review-fixes block, which we copied word for word into both families' `stage.py`. The contract checker reports 1 row, 0 problems, and when we re-run `stage.py` it reproduces both files byte for byte. We re-opened no primary. +- On 3 November, the contract checker reported 1 row, 0 problems, and when we re-ran `stage.py` it reproduced `manifest.jsonl` and `claims_staged.jsonl.gz` byte for byte. We re-opened no primary, because the session's network policy refused trussell.org.uk. +- On 29 October, while the row stood in `uk_thinktank_misc`, the lane re-opened the saved page at the locator and checked the quote and value against it. It also checked that the quote runs as contiguous text under 15 words and contains `value_raw`.