[juaristi22]

Summary

PR2 of the plan for the November event data (microcosm#1095). It re-pins the UK feed onto Chronicle main after PolicyEngine/chronicle#423, #424 and #425, which bring PolicyEngine/chronicle#420's facts for the November UK events, and carries the pieces that read them. It stays on policyengine-uk 2.174.1, as main locks it.

Closes #1439 (T1). That issue's last "Done when" item, the national rebuilt and the trading allowance measure rescored on it, follows with PR2's datasets after the merge.

Draft. María signed the PR's texts on 9 November 2026 (see "Rulings and signatures"). Lane P's PIP points stage (P1) and Y1, the admin-linked limited capability status, are in, and María signed their eleven choices the same day. The arm at the signed head and four calibrate-only variants ran on 9 and 10 November (see "Measurement"). On their results María chose on 10 November to bind Y4's two age bands and P2's PIP award split, and to keep S4 held out. She signed the one restated register text the same day.

• The re-pin to 8045b13 (2,091,616 rows).

  • A row-by-row diff of the two artifacts finds no row dropped and no value moved; 506,321 rows are new.
  • No compiled value moves on either surface: both reference files and both compile-parity receipts are byte-identical. The vendored resources carry the same rows under the new feed identity.

• Nit 3 (Vahid's review of UK November event data, PR1: full-time education, single ages 16 to 24, Scottish PRIS diagnostics, energy kWh columns #1535): apprentices only outside full-time education.

  • frs_person_draws no longer flags an employee aged 21 to 24 who is in full-time education (is_in_full_time_education). The FRS never codes an apprentice that way.
  • The draw's target is unchanged; only its pool narrows. On 2.174.1 only the s.9B employer NICs relief reads the flag at these ages.

• T1, microcosm#1439: the trading allowance claim status in three outcomes.

  • trading_allowance_receipts writes trading_allowance_claim_status: CLAIMANT for drawn claimants, NON_CLAIMANT for other traders with a positive profit, UNKNOWN otherwise. self_employment_gross_receipts stays at the engine's zero.
  • The full-relief group (FRS respondents with a profit of £1,000 or less, not drawn claimants) is empty at its share of zero. The five bound SPI self-employment rows count declarers only, so no row moves.
  • The claimant rates are re-fitted on control C2's release to HMRC's figures of about 400,000 gainers at about £380: 0.6364 for profits up to £4,500 and 0.1532 up to £9,000 (from 0.72 and 0.14).
  • tools/measure_uk_trading_allowance_gainers.py judges the rates on the expected gainers, each band's rate times its would-be gainers, against 400,000 within 5%. The release's realised count is reported beside it (María's ruling of 9 November 2026).

• S4: HMRC's Scottish NSND outturn 2024-25, two held-out diagnostics. These are the revenue (£18,635m, uprated to 2025 by the OBR earnings index) and the taxpayer count (3,036,400, read at 2025 under a reviewed hold, since no pinned index counts taxpayers).

• S2(a): Scotland's council tax bills levelled.

  • The level is the Scottish Government's 2024-25 liability before council tax reduction: CTAXBASE 2025's Band D equivalents at September 2024, times CTAS 2026's average Band D, over the 32 councils, £3.393bn.
  • One factor at the spine design weights scales the bills. The bound band stock keeps the mix, so no weight moves.

• Y4: DWP's people on Universal Credit aged 16 to 24, 16 to 21 and 22 to 24.

  • Great Britain, every regime, on the four quarterly points of 2025.
  • 16 to 21 and 22 to 24 bind (María's call of 10 November 2026). 16 to 24, their sum, stays held out as the check on them.
  • The 16 to 17 count never appears alone or by difference.

• P3: DWP's PIP cases with entitlement, 94 rows. England and Wales are separate rows, each with its total, the awards of each component and age bands; the nine English regions are split by each award. All are single breakdowns. Each country's six award rows bind, the award split (P2, María's call of 10 November 2026); the other 82 stay held out.

• Y3: ONS's NEET people aged 16 to 24, held out. It is the mean of 2025's four LFS quarters, not seasonally adjusted, on a new quarterly_window_average compile rule.

• The UC take-up redraw fits young claimants by age (uc_take_up, María's call of 9 November 2026).

  • Y2's flag takes full-time students off UC, and the redraw refilled the single caseload by award band without reading age.
  • Now, within every pool, the drawn units with a claimant or partner aged 16 to 24 claim at one odds ratio against the pool's other drawn units. The ratio is solved so that young claimants and partners come as near DWP's count as the redraw allows.
  • On the first arm it stopped at the low end of its range (see "Measurement"). So, on María's second call of 9 November 2026, a shortfall passed between pools now fills only units with no young claimant or partner. A family type whose shortfall only young units could fill falls short of DWP's count by it, and the receipt records what stays unplaced.

• P1: PIP points for every recipient (pip_points, María's go of 9 November 2026).

  • policyengine-uk decides a PIP component's rate from pip_dl_points and pip_m_points (8 points standard, 12 enhanced) wherever no category is supplied for the year. The release used to store the categories, which the engine copies into every projected year, so no reform to the thresholds could move anyone.
  • A new late stage draws each recipient's daily living and mobility points within their stored award, by age group. The source is DWP's Pathways to Work evidence pack, chapter 2, Tables 2.9 and 2.10 (England and Wales, end of January 2025; chronicle#420 item 5).
  • The release carries the points instead of the categories, and the spine keeps the categories for VED and the severe disability flag. With unchanged thresholds the engine's categories equal the stored ones for every person, which an engine test checks for 2024 to 2030.
  • Table 2.9's whole percentages add to between 92 and 102 per row as ported (16 to 19 is the 92). The draw uses each row's shares only relative to one another within an award's range, and the receipt records every row's total.

• Y1: DWP's limited capability status on linked UC claims (uc_limited_capability, María's NS6 ruling of 9 November 2026, which revisits her ruling of 29 October to keep the engine's proxy).

  • policyengine-uk gives limited capability for work and work-related activity to anyone flagged is_disabled_for_benefits (any reported disability benefit or ESA), unless their own ESA award lacks the support component.
  • The FRS 2024-25 links most UC claims to DWP's records, and BENEFITS.UCLCW gives the claim's own status (SN 9563's data dictionary: 0 None, 1 LCFW Group, 2 LCFWRA Group; blank where the claim isn't linked). frs_spine carries it on the UC row's person; the FRS keeps one UC row per claim, couples included.
  • A new engine-backed stage just before uc_take_up stores both flags for everyone: the locked engine's proxy, with DWP's status on linked FRS claims. Unlinked reporters, the take-up redraw's entrants, the legacy-closure movers and every SPI row keep the proxy.
  • On the age-fit arm's spine, the status and the proxy disagree on a substantial share of linked claims in both directions, so Y1 changes who has the LCWRA element more than how many. Before take-up, the model's LCWRA units already sit near DWP's count, so the bound LCWRA row's starting gap looks like a take-up question rather than a flags one.

Measurement

• Control C2 (0dff7773e: main after PR1 plus the re-pin; review date 2026-11-07; feed 8045b13):
  • all seven terminal gates pass, with 98.7% of 1,514 rows within 10%;
  • its release is identical to main's last national build, every column and every weight.
• The first arm (this branch's data at cf73a73ad, before the squash; review date 2026-11-09):
  • all seven terminal gates pass. The final loss is 0.00515 (C2 0.00514), with 98.8% of 1,514 rows within 10%.
  • The age fit stopped short, which led to the age-aware gap rule (see "Summary"). At its smallest ratio, young claimants and partners still came to 1.24 times DWP's count at the stage's weights. A replay of the stage on the spine's checkpoint (identical receipt and output) showed why:
    • about half the young claims are units that report UC, which no ratio moves;
    • almost all the rest sat in single award-band pools whose count needed every drawn unit, mostly the £200 to £400 a month bands around the under-25 standard allowance;
    • those pools were full because the shortfall of DWP's high-award single counts passed down to them: the model entitles fewer high-award singles than DWP counts.
  • Held out on its release: Y4's 16 to 24 read 1.28 times DWP's 762,585; Y3 was 2.7% below ONS; England's PIP caseload was 3.9% above DWP's and Wales's 0.4%, with England's cases aged 16 to 34 about 42% above.
• The arm at the signed head (daa359a0b; a full spine, since P1 and Y1 change stages; review date 2026-11-09):
  • all seven terminal gates pass. The final loss is 0.00515, with 1,497 of 1,514 rows within 10% (98.9%).
  • At the spine: at the spine's design weights, Y4's 16 to 24 row meets DWP's count (1.002 times), so the age-aware rule works. A replay of uc_take_up reproduces the stored receipt. Replayed with and without Y1's flags, it gives 1.2% more Great Britain benefit units claiming with an LCWRA element with them.
  • P1: on the release, the engine's categories from the stored points equal the stored awards for every recipient.
  • T1: 399,300 expected gainers (−0.2%), within the 5% tolerance; the release's own draw gives 379,100.
  • The bound UC rows start lower at the spine's weights than on the first arm (households −11.2% against −6.7%; households with a deduction −15.8% against −11.1%), and end within a point of where they ended there.
  • Held out on the release:
    • Y4's 16 to 24 reads 1.24 times DWP's count, so the solve puts back nearly all that the age-aware rule took off at the spine.
    • Y3's NEET count moves from 2.7% below ONS to 16.2% above. The proxy's count at design weights is the same on both arms, so the move is all weights: the solve refills the bound UC rows through the young people still drawn to claim, mostly NEET.
    • P3: 33 of 94 rows within 10%.
• Calibrate-only variants on that arm's spine (the same review date; each a local commit on the signed head, never pushed):
  • S4 (both NSND rows bound): both land within 0.3% and all seven gates pass, but five SPI income-band rows cross 10%, four of them Scottish. The solve can't lower Scotland's NSND tax by about 5% while SPI's tax on all income stays bound by band, so it trims Scottish taxpayers in the middle bands. S4 stays held out.
  • P2 (England's and Wales's award split, 12 rows): the 12 rows land within 0.5%. One Scottish SPI row that sits on the 10% line from arm to arm crosses it.
  • Y4 (16 to 21 and 22 to 24): no row crosses 10%, and 16 to 24 lands at −0.1% as their sum. NEET falls to 3.1% above ONS. P3: 36 of 94 within 10%; England's cases aged 20 to 24 go from 37% to 14% above DWP's.
  • Y4 and P2 together (the register this PR now carries):
    • all seven terminal gates pass. The final loss is 0.00513, with 1,510 of 1,528 rows within 10%.
    • All 14 bound rows land within 0.3%, and 16 to 24 lands at +0.1%. NEET is 1.1% above ONS.
    • England's and Wales's PIP totals land (0.0% and −0.1%), since each component's awards sum to the caseload.
    • P3's other 82 rows: their mean absolute error falls from 0.226 on the base arm (0.220 with Y4 alone) to 0.203, but 24 are within 10%, against 27 (29). England's young bands come closer: 16 to 19 is 37% above DWP's (52% on the base arm), 20 to 24 is 7% above (37%) and 30 to 34 is 26% above (37%). With the totals pinned, the older bands slip 1 to 4 points further below than with Y4 alone, so the caseload's age profile still runs young.
    • Elsewhere: two bound ONS population rows (women aged 19, and ages 20 to 29 in the South West) move by 3 to 4 points and stay inside 10%. One SPI income-tax band row that sits on the 10% line crosses it.

Rulings and signatures

María's call of 10 November 2026, on the calibrate-only variants: bind Y4's 16 to 21 and 22 to 24 rows and P2's award split, and keep S4's two rows held out (b835be176; the signature below is 33c273773).

• The two Y4 parts' exclusions and the twelve award rows' exclusions go: the register holds 135 rows, from 149.
• Signed by María on 10 November 2026: 16 to 24's restated reason and adjudication. The reason now says the solve binds its two parts and it stays held out as their check; the adjudication records the call. The entry carries approved_on 2026-11-10 and expires 2027-02-10, so the next arm needs a review date from 2026-11-10.

Signed by María on 9 November 2026. The registers carry approved_on 2026-11-09: the exclusions expire 2027-02-09 and the S4 hold 2027-01-09. The arm's review date is the signing day.

• Nit 3: the restated frs_person_draws eligibility text.
• T1:
  • the full-relief group's rule and its share of zero;
  • the declarer-only filter on the five SPI self-employment rows;
  • the restated SPI donor savings rebased to SPI's 2023-24 bands (savings follow-up to #1418) #1428 filer stance in the stage notes;
  • self_employment_gross_receipts re-signed at zero;
  • the carrier kept on the export, and all five rows filtered;
  • the re-fitted rates.
• S4: the two exclusions and the taxpayer count's reviewed_no_index uprating hold.
• S2(a):
  • the level's rule;
  • one factor rather than one per band;
  • the September 2024 base;
  • a frame with no Scottish bill keeps its bills.
• Y4: the three exclusions. Their reason states what the age fit can and can't do. Two go on 10 November, and 16 to 24's is restated (above).
• P3: the 94 exclusions, of which 82 stay after 10 November.
• Y3: the exclusion, with two choices within María's rule: not seasonally adjusted, and calendar 2025's quarters.
• P1 (pip_points):
  • Points within a band: spread evenly over its integers, with the open band (41 and over) at 41.
  • Scotland's and Northern Ireland's recipients use England and Wales's tables.
  • Pension-age recipients take the 60-and-over row, which carries them.
  • The draws are seeded on person_id.
  • The two categories leave the release. This covers their reviewed export exclusions and the coverage manifest's two engine-derived entries, dated 2026-11-09. The manifest tool admits those entries through a reviewed list of inputs the pinned engine derives since the eFRS reference's 2.89.0.
• Y1 (uc_limited_capability, NS6 revisited):
  • NS6 restated: read UCLCW for linked claims, and keep the proxy elsewhere. The bound LCWRA row's note records it, and no longer says the carer and childcare rows sit on the measure-exclusion register, since neither does.
  • The code mapping: 0 None, 1 LCW, 2 LCWRA; blank means not linked.
  • The joint-claim rule: where the proxy flags no claimant, the status goes to the UC row's person, or to the partner where the UC row's person cares and the partner doesn't, so a carer keeps the carer element.
  • NONE overrides the proxy: a linked claim with no status has no LCW or LCWRA, whatever its disability benefits.
  • Who keeps the proxy: unlinked reporters, the take-up redraw's entrants, the legacy-closure movers and SPI rows. uc_reporter_redraw's award screen, earlier in the spine, still reads the proxy.

Still to come in this PR

• Review rounds.

Not in this PR

• P2's original award-mix correction: P3's data put the gap in the age profile, not the award mix, so this PR binds the award split instead.
• PIP's age profile: it still runs young with the split bound (see "Measurement"), and nothing here corrects it.
• Y4's split by conditionality regime: it waits for María's mapping of DWP's regimes.
• P4, binding Scottish ADP and Carer Support Payment spending: a limitation instead. Scottish disability and carer payments (Adult Disability Payment, Carer Support Payment) are not calibrated to Social Security Scotland: the data hold about a third fewer Scottish recipients than its caseload, and binding the spending would stretch those households' weights beyond the build's limit.
• The policyengine-uk bump.

Derived surfaces

• Regenerated on the national cut of feed 8045b13:
  • the national target references and membership: 1,697 active references, the 1,599 before plus the 98 new rows (84 still held out after the binds), with no existing reference moved;
  • the national compile-parity receipt, with 98 new ledger-only rows;
  • the release-input coverage manifest;
  • the H2 spine fixture.
• The local surfaces move only with the re-pin's feed identity.
• Three new vendored resources come from the pinned feed: scotgov_band_d_liability_facts.json (S2(a)), dwp_uc_people_by_age_facts.json (the age fit) and dwp_pip_points_facts.json (P1).
• For P1:
  • the coverage manifest's pip_points family, and the two categories as reviewed exclusions (141 required and 4 reviewed of 145 inputs);
  • the H2 fixture's 55th stage;
  • the release contract's committed digests in microcosm-data (the spec's policy, gates manifest and fingerprint, and the release cut's scoped parts), re-pinned on this spec. They also catch up with PR2's earlier commits, whose spec and gates.json edits had left them at main's values.
• For Y1:
  • the coverage manifest's uc_limited_capability family (still 141 required and 4 reviewed of 145 inputs);
  • the H2 fixture's 56th stage, and UCLCW in its synthetic BENEFITS rows;
  • the release contract's digests, re-pinned again.
• For the binds: none move. No committed reference, receipt, fixture or digest reads the measure-exclusion register or the contract's parity notes.

Testing

Targeted tests by touched surface at each step, one process per file:

• The re-pin:
  • 14 direct files: 444 passed;
  • a sweep of 55 engine-free files: 2,481 passed, 11 skipped;
  • H2 parity: passed.
• Nit 3, T1, S4 and S2(a):
  • 43 engine-free files: 2,433 passed, 10 skipped;
  • 8 engine-tier files: all passed, including T1's claim status (9) and H2 parity.
• Y4, P3 and Y3:
  • the shared compiler suite: 514 passed;
  • after the regeneration, the 30 files that read the references: all passed.
• The age fit:
  • 21 engine-free files: 1,243 passed, 1 skipped;
  • H2 parity: passed.
• The final text corrections:
  • the 19 engine-free files that read the exclusions register, the stage module or the changelog: 1,124 passed, 6 skipped;
  • H2 parity: passed.
• The age-aware gap rule, at the head: 23 engine-free files that read the stage, the spec or the registers: 1,400 passed, 1 skipped. H2 parity, on the regenerated fixture, passed. On the stage's saved inputs from the arm, the committed code reproduces the replay's result exactly.
• The signing commit:
  • 32 engine-free files that read the registers, the stage spec, the coverage manifest or the contract digests: 1,894 passed, 6 skipped;
  • H2 parity, on the regenerated fixture: passed.
• P1, at the head:
  • the full set a stage change reaches, 302 engine-free files (microcosm-build's 274 UK files, the 18 spec-engine files, the gate-battery, coverage and spec files, and microcosm-data's contract test) and 8 engine files: 8,211 passed, 39 skipped and 8 failed of 8,258;
    • two failures were P1's own pins in the engine test_uk_release_input_coverage.py, now restated;
    • two were tests T1 and Nit 3 had left stale, test_uk_pension_age_benefit_targets and test_uk_stochastic_tools, now pinned in e7f8223a3;
    • four fail only on the local machine: test_uk_cgt_projection (the engine is installed there), the two in test_uk_rowwise_build_driver (a temp folder under home prints as ~; they pass outside it) and test_uk_orrery_reconstruction (Orrery's Node package isn't installed locally);
  • the restated and pinned files, rerun: all passed;
  • the 13 other engine files that read the stage list, the registers or the coverage ledger: 117 passed;
  • H2 parity, on the regenerated fixture: passed.
• CI at a859ac7c9 failed in the engine-free job, in test_uk_release_assembler, on the release contract's stale digests, which 2e0dc5771 re-pins. CI stops at its first failure.
• Y1, at the head (targeted tests, as María asked: the files Y1's surface reaches):
  • 26 engine-free files (the spine and its carrier, the new stage, the roster and spec files, the export, coverage and contract registers, the LCWRA row's readers, uc_take_up and the concept mapping) and 7 engine files: 1,704 passed, 5 skipped and 2 failed of 1,711;
    • both failures were pins Y1 moves: uc_take_up's preceding stage, and the engine coverage test's required build stages. Restated; the two files rerun pass (34 and 52);
  • the stage's own tests: 6 engine-free, and 2 engine on 2.174.1 (unlinked people keep the engine's proxy exactly, and the LCWRA element follows DWP's status in 2024 and 2025);
  • H2 parity, on the regenerated fixture: passed.
• The signing commit (daa359a0b): the 13 engine-free and 2 engine files that read the signed texts, the coverage manifest or the contract: 920 passed and 5 skipped of 925. H2 parity, on the regenerated fixture: passed.
• The binds (b835be176), the files that read the measure-exclusion register or the contract's notes:
  • 26 engine-free files: 1,206 passed, 9 skipped and 1 failed of 1,216. The failure was the register's count pin in test_uk_measure_simulation, now 135; the file reruns pass (65);
  • the 2 engine files that key the target kernels on the register, and test_uk_release_assembler: 38 passed.
• The signature (33c273773): the 20 engine-free files that read the measure-exclusion register and the 2 engine files above: 1,056 passed, 1 skipped.
• CI at 33c273773 failed in the engine-free job (3.14) on one test of 22,242: a stage-order pin in test_uk_legacy_closure_take_up, stale since Y1 put uc_limited_capability before uc_take_up. 09add3214 restates it, and the file passes (26).
• UK trading allowance: store the claim status once policyengine-uk#2765 is released #1439's allowance above £3,000 (e4185628d): a reform to £5,000 in 2028-29 relieves a claimant's 2025 profit by min(profit, £4,000) and a non-claimant's by nothing. T1's engine file passes (11).

Checks:

• ruff check and ruff format --check are clean on the changed Python files, P1's 24, Y1's 22 and the binds' 3 included.
• tools/ci_test_plan.py verify passes.
• The feed-gated tests weren't run locally; CI skips them too.

Disclosure

Figures here are national, Great Britain, England, Wales or Scotland-level aggregates, and one English region's population row, each resting on 30 or more survey households. Statements over several rows include cells on fewer: the twelve bound award rows (Wales's nil cells rest on 10 to 29 households) and the counts and averages over P3's rows (cells on 3 to 29 households among them). No figure rests on fewer than three. They include no unit-record values and quote no SPI-channel rows.

🤖 Generated with Claude Code

[vahid-ahmadi]

Review of 8780aa3

I read the code and config changes against the merge of #1535 (dc00d15): the re-pin, Nit 3, T1, S2(a), S4, Y4, P1, Y1 and the take-up age fit. I spot-checked the generated files against the re-pin, and checked the law-facing claims against legislation.gov.uk, policyengine-uk 2.174.1 and HMRC's 2024-25 Scottish outturn release. In a fresh clone I ran the PR's engine-free tests one file at a time: test_uk_pip_points, test_uk_uc_limited_capability, test_uk_frs_council_tax, test_uk_trading_allowance_receipts, test_uk_trading_allowance_gainers_tool, test_uk_apprentice_draw, test_uk_uc_take_up, test_ledger_quarterly_window, test_uk_chronicle_feed, test_uk_uprating_holds and test_uk_legacy_closure_take_up. All passed.

From #1535: all four items are closed. Should 1 was counted, with fewer than three households, and noted in the stage notes. Should 2 was fixed in ab629d689. Nit 4 went into 821c13d. Nit 3 is here at frs_person_draws.py:328, and the stage now refuses a frame without is_in_full_time_education (:303).

Should

1. A PIP points-threshold reform will also move Scotland's Adult Disability Payment recipients, and Northern Ireland's PIP recipients (uk_runtime/pip_points.py:30-34 and :153-159; frs_spine.py:175-176). frs_spine maps ADP's FRS codes 117 and 118 into pip_dl_reported and pip_m_reported. So Scottish ADP recipients hold pip_*_category, and this stage now gives them points. policyengine-uk 2.174.1 has one threshold for everyone, gov.dwp.pip.*.points_thresholds, and no separate ADP variable. Failure case: a Scottish ADP standard-rate recipient drawn at 8 daily living points loses the award under a reform that raises daily_living.points_thresholds.standard to 9. ADP's thresholds, though, are set by the Disability Assistance for Working Age People (Scotland) Regulations 2022 reg. 8(3) (8 to 11 points limited, 12 or more severely limited), and a DWP reform doesn't change them. NI PIP sits under its own 2016 regulations. Before the stage, the stored categories meant no reform moved anyone, so this is new with P1. Suggested fix: say it in the changelog and stage notes ("a threshold reform reaches ADP and NI recipients too; net them out by country"), and add each country's recipient count to the receipt so users can do that. A region gate on the thresholds would be a policyengine-uk change, not one for this PR.

2. Scotland's level counts dwellings no FRS household lives in (uk_runtime/frs_council_tax.py:323; changelog.d/1095-uk-scottish-council-tax-level.changed.md:3). The level is CTAXBASE's Band D equivalents times CTAS's Band D, £3.393bn (I recomputed it from scotgov_band_d_liability_facts.json: 2,392,465 Band D equivalents over 32 councils). The vendored rows are CTAXBASE 2025's "Band D equivalents" table, with chargeable dwellings as their entity. gov.scot's "Second homes and empty properties in September 2025" comes from the same council tax base returns. It counts 20,927 second homes and 44,453 empty homes liable for council tax in September 2025, after a fall of 679 and a rise of 915 on the year. That puts September 2024 at about 21,600 and 43,500. The FRS only samples occupied households, so one factor over the FRS bills puts the liability on these dwellings onto households. Failure case: about 65,000 liable dwellings with no FRS household is roughly 2 to 3% of Scotland's chargeable stock. If they pay about an average bill, or more where a premium applies, the factor runs high by about that much, and so does every Scottish household's council_tax. That flows into CTR entitlement and household net income. Suggested fix: take second homes and liable empty dwellings out of the level, from CTAXBASE's own counts by discount or premium class at September 2024. If that can't be done in this PR, record in the stage notes and changelog how much of the roughly 9% gap they could account for.

Nit

3. The PIP points are totals only, drawn independently by component (pip_points.py:20-27). The stage can score a change to the 8 or 12 thresholds. It can't score a rule that turns on points in one activity, such as needing at least 4 points in one daily living activity, and it loses any link between a person's daily living and mobility points. One sentence in the changelog's scope would stop users reading more into the points than they hold.
4. Once its parts bind, 16 to 24 carries no new information after the solve (calibration_measure_exclusions.json:546). It lands where they do (+0.1%). The reading that tests the age fit is the one at the spine's design weights (1.002), so it is worth reporting that one on every arm.

Checked and fine

• Re-pin: chronicle_feed.json moves to 8045b13 (2,091,616 rows). In every vendored resource the 38 4/4 files change only source_commit, fact_row_count, facts_sha256 and manifest_sha256. In desnz_constituency_energy_facts.json only the derived keys and the input hash change, with no value moved. The local membership and census change the same way. No 64753803 remains in the packages or tools.
• T1 against ITTOIA 2005 Part 7A: full relief applies without an election where relevant income is within the allowance (s. 783AE(1)), and trade profits are then nil (s. 783AF). Partial relief deducts the allowance in place of expenses (s. 783AI(3)). So a CLAIMANT's receipts are their profit plus £1,000, and a reform to £5,000 relieves min(profit, £4,000), as e4185628d tests. The re-fit solves exactly from fit-ctl2-0dff7.json: 0.6364 × 539,000 + 0.1532 × 372,200 = 400,041 gainers, and 0.6364 × £196.8m + 0.1532 × £174.6m = £152.0m, which is £380 a gainer.
• S4: HMRC's 2024-25 outturn gives £18,635m of NSND liability and 3,036,400 Scottish taxpayers. HMRC defines taxpayers as people with some net NSND liability after reliefs, so the count fits the model's taxpayers.
• P1 thresholds: PIP Regs 2013 reg. 5(3) and reg. 6(3) set 8 and 12 points. PIP_AWARD_RANGES (pip_points.py:102) matches, and the mobility totals (0, 4, 8, 10 to 24 even) are the sums the descriptors allow. policyengine-uk 2.174.1's utils/pip.py derives the category from the latest points only where no category is supplied, so dropping the categories from the export is what makes the points bite.
• Y1: the proxy is policyengine-uk 2.174.1's own formula (UC Regs 2013 regs 39 and 40), and is_disabled_for_benefits is a plain input there, so storing both flags drops no formula a reform could reach. The joint-claim choice keeps a carer's carer element, which reg. 29(4) would otherwise bar for a claimant with LCWRA.
• The age fit: the target is DWP's Great Britain claimants and partners aged 16 to 24, all regimes, averaged over 2025's four quarters. I summed February 2025's single years from the vendored facts: 749,143.

CI at 8780aa3: lint, wheels, installed-emitter, select-countries, integration-uk, engine-uk (3.13) and engine-uk (3.14) pass. engine-free (3.13, 3.14) and engine-us (3.13, 3.14) are still running.

Verdict: the code matches its signed texts and the law I checked. Should 1 and Should 2 are the two to settle before the national is rebuilt on it. Both can be handled with notes if the fixes wait.

🤖 Generated with Claude Code

[juaristi22]

Thanks, Vahid.

Should 1 (ADP and NI PIP under a threshold reform): agreed. a0c6d8b34 adds each UK country's recipients to the stage's receipt (rounded records, a support band and the weighted count, withheld below three source households). The P1 changelog now says that a threshold reform reaches Scotland's ADP recipients and Northern Ireland's PIP recipients, whose awards follow their own regulations, and to net them out by country. The stage notes say it too, signed by María (b4250ebe9). A region gate on the thresholds stays a policyengine-uk question.

Should 2 (second homes and empty dwellings in Scotland's level): agreed: the level carries their liability. gov.scot's Second Homes and Empty Properties in September 2025 puts September 2024 at about 21,600 second homes and 43,500 liable empty homes, roughly 2 to 3% of the chargeable stock. The stage notes and the changelog now say how much of the level they could carry, signed by María (b4250ebe9). Netting them out needs CTAXBASE's counts by class in Chronicle, so it follows separately.

Nit 3: the P1 changelog's scope now says the points are each component's total, drawn independently. They score a change to the 8 or 12 thresholds, not a rule on one activity's points, and they keep no link between the components.

Nit 4: agreed. The arm reads take Y4's 16–24 at the spine's design weights beside the release; for the head arm those are 1.002 and +0.1%. We'll report the spine reading on every arm.

🤖 Generated with Claude Code

[vahid-ahmadi]

Round 2 at b4250eb

I read a0c6d8b34 and b4250ebe9 against 8780aa3 and María's reply, and re-ran test_uk_pip_points in a fresh clone (9 passed).

1. Should 1 (ADP and NI under a threshold reform): fixed. The receipt now has recipients_by_country (uk_runtime/pip_points.py:171-186, :523-554, :646), and the changelog says to net them out (changelog.d/1095-uk-pip-points-stage.added.md:18). I checked the counts. They key on recipient, which is any person with a category other than NONE. frs_spine.py:175-176 merges codes 96/97 (PIP) with 117/118 (ADP), so Scotland's row is every Scottish recipient under either code. Northern Ireland has its own key. The 12 region names match frs_spine.py's REGION_MAP (:101-114), and an UNKNOWN region falls into "unassigned" rather than into a country. The weighted count is withheld below three source households, and the new test covers that. One small thing: the counts are at the stage's spine weights. A user netting out at release weights should filter on region in the release, not lift these numbers. A short note in the receipt saying so would help.
2. Should 2 (second homes and empty dwellings in Scotland's level): partly. Notes only, no code change (changelog.d/1095-uk-scottish-council-tax-level.changed.md:8-11; the frs_council_tax stage notes). The figures are right: about 65,100 liable dwellings is roughly 2.6% of the stock, and up to about 5% once premiums count. But I don't think a dwelling-count correction has to wait for CTAXBASE's class counts in Chronicle. D14f, María's ruling of 2026-10-29, already settles that these dwellings don't belong in main-residence bills. obr.council_tax binds Great Britain's council_tax_less_benefit, Scotland included, scaled by England's chargeable dwellings less mhclg.council_taxbase.second_homes and mhclg.council_taxbase.empty_dwellings (uk/uk_population_targets.json:2932, complement operand at :2968). The note there says "England's share stands for Scotland's and Wales's" as a stated approximation, and says binding the whole line "pushed main-residence bills up". S2(a), though, levels Scottish bills to all chargeable dwellings (frs_council_tax.py:323). So one build now treats the same dwellings two ways. The bound target nets them out, and the Scottish level puts them back into each Scottish household's council_tax. The share's inputs are already in the pinned feed. Only the premium refinement needs class counts, and I found no Scottish second-home concept vendored yet. Suggested fix: scale level.total by the D14f share, stated as the same approximation. Or say in the S2(a) notes why the level departs from D14f. Either way, the "follow-up" should name an issue. Households' council tax on second homes and other property (engine and data) #1423 covers council tax on other property, not this.
3. Nit 3 (points are totals only): fixed (changelog.d/1095-uk-pip-points-stage.added.md:19; stage notes and the pip_points.py docstring).
4. Nit 4 (16 to 24 at design weights): accepted. It is a reporting change for the arm reads, so there is no code to check here.

Records: text A sits in sources.yaml, source_stages.json, the parity fixture, the pip_points.py docstring and the P1 changelog. Text B sits in sources.yaml, source_stages.json, the fixture and the S2(a) changelog, all dated 2026-11-10. That matches what the commit message says. The coverage manifest changes only source_manifest_sha256.

CI at b4250eb: lint, wheels, installed-emitter, select-countries, integration-uk, engine-uk (3.13, 3.14), engine-us (3.13, 3.14) and engine-free (3.14) pass. engine-free (3.13) is still pending.

Verdict: Should 1 and both nits are done. Should 2 is the one left. Its notes are accurate, but the D14f share already gives an interim correction, so the deferral needs either that or a reason.

🤖 Generated with Claude Code

[juaristi22]

Thanks, Vahid.

1. Should 1: agreed on the note. The counts are at the stage's spine weights; to net ADP and Northern Ireland recipients out at release weights, filter on region in the release. We'll add that line to the receipt in PR3 of the November event-data plan rather than hold this PR for it.

2. Should 2: we'll defer the correction to PR3 of the November event-data plan (#1095), which nets these dwellings out of the level. You're right that D14f already nets them out of obr.council_tax, while S2(a) keeps them for Scotland. We're not applying D14f's share in this PR for three reasons:

• An unmeasured move. It would lower every Scottish household's council_tax by about 2.6% (up to about 5% with premiums) after the arms that measured S2(a) as signed, so it needs an arm of its own. PR3 has one.
• England's share for Scotland's. It would layer England's share onto a level that is otherwise Scotland's own. Scotland's counts, and its premiums since April 2024, can differ from England's, and gov.scot warns that its empty-home counts may be overstated. PR3 uses CTAXBASE's own counts by class once they're in Chronicle, with D14f's England share as the interim if they can't be packaged in time.
• The size meanwhile. On the Great Britain line that obr.council_tax binds, the difference is about 0.2 to 0.4%. The S2(a) notes and changelog already state the dwellings and their size.

3 and 4: thanks.

🤖 Generated with Claude Code