Skip to content

Rewrite the Autumn Budget 2026 harvest's NOTES.md files in active voice - #246

Merged
MaxGhenis merged 66 commits into
mainfrom
uk/ab2026-notes-voice
Nov 5, 2026
Merged

MaxGhenis merged 66 commits into
mainfrom
uk/ab2026-notes-voice

Conversation

@MaxGhenis

@MaxGhenis MaxGhenis commented Nov 5, 2026 •

Copy link
Copy Markdown
Contributor

We rewrite the 12 NOTES.md files under sources/harvest-uk-ab2026-2026-10-29/ in active voice and keep every fact. We branched from #231, which merged as f6edbdf6, and retargeted this PR to main, as part of #230.

We use #247's keys and row counts and describe its ingest outcomes, so we merged #247 first, as 1615e267. In #247 we registered the 86 keys we proposed and moved 973 rows' conditions.proposed_measure_key into measure_key. We also renamed the welfare fraud key to ab2026__package_welfare_fraud_and_error_savings_in_forecast. A script computed each family's key and row counts from #247's claims_staged.jsonl.gz at 3812ce4d, and we get the same rows and counts at a28223f2. When we record what we did on a date, we keep the date and past tense.

  • We change only the 12 NOTES.md files, and leave claims files, COUNTS.json, manifests and the README as they are.
  • A local check, which we didn't commit, compares numbers, dates, backticked spans, quoted phrases, URLs, hashes, fenced blocks and ab2026 keys between each file's old text (a9f81e87), the first rewrite (fbe0d76c) and this text. We give its result per family in the hand-off comments.

Changes in answer to the review of fbe0d76c

The reviewers requested changes, and we made them family by family, one commit each:

  • Facts we moved in the first rewrite (findings 1–23). We restored the actual-ANI restriction in CenTax, the inverted chart-label logic and comment reference in RF, and "a single Table 6.1 line" in TPA. We also restored the readers' 4 November edit of TPA's article and the OBR rows in the rule 10 count. We attribute the devolved-statement readings to ourselves and the first lane's 29 October IFS choices to that lane in the past tense. The coordinator assigned the four thinktanks-b URLs in the staging-branch brief, and we now cite that brief in the notes.
  • The writing gate (findings 24–33). We cut the self-referential stage.py gloss and assigned actions to the people who performed them. We rewrote the proposed-key table cells and the other cells the reviewers named, cut label openers and defensive clauses, and joined clipped runs.
  • The nine flagged sentences (B1–B9). We checked all nine sentences against the rows or repository. We use Add CenTax's costing of removing the £100,000 childcare limit #207's parents_bunching name, date the £4.4bn with rows 34, 54 and 127, and own the judgement "very probably". For "Figures 3.51", we refer to paragraph 3.51's two figures.
  • Statements that would go stale once Ingest the Autumn Budget 2026 harvest (tranche 2 of #230) #247 landed (section D). We renamed the "Proposed measure keys" headings, which no reviewer on uk/ab2026-harvest-staging cites, and kept the records of keys we withdrew or retired.

Changes in answer to the review of 4a34051b

We call the keys we proposed "keys we proposed" when we list single measures, and attribute the Barnett total and component readings to ourselves. We fix the RF dateModified and OBR social care sentences and keep the first IFS lane's record in its own voice. We tie TPA's 4 November check to review B's fetch and split the long value lists into sub-bullets.

Changes in answer to the review of b2347435

We corrected three IFS passages against the rows. For the rows we kept under the 3 November addendum, we now state Boileau's metrics on 21 rows and the signs and scalings on Adam's 12. We also give Adam's key counts twice, for the first lane's 18 rows and for the 12 we stage. We gave the first lane the past tense on Wernham's six dropped metrics, named the actors the reviewer listed, cut "explicitly" where it only added emphasis, and limited the OBR's social care sentence to what the rows show.

Changes in answer to the review of 1ad235f3

We corrected two statements. In RF's notes we gave README rule 10 as the reason we move OBR row 2740 to revenue_change, but under that rule paragraph 3.43 counts as a stated effect on borrowing. We then gave the reason the ingest recorded, that the Government didn't adopt the freeze. In the OBR notes we now point to row 2740, not row 2737, for the continued freeze's effect on borrowing.

#247's ingest now scores IFS's £80 on average_household_income_change and tallies its £140, £24 and £36, and we state that in the IFS notes. We also named the actor in the sentences the reviewer listed, searched the 12 files for rows, tables, pages and documents acting as subjects, and rewrote what we found outside quotes.

Changes in answer to the review of 69430330

We restored five minus signs in the OBR notes on savings and property income tax rates: the OBR prints −0.1 and −0.9 in Table 3.1 and −2, −543 and −400 in Table 3.3, with positive values for borrowing increases. We also split that sentence and the first on Other AME spending decisions, which ran over 40 words.

#247's ingest now keeps OBR row 2740 on decisions_effect_on_borrowing under the macro verdict, so we corrected RF's two sentences that moved it to revenue_change.

The reviewer named ten more sentences with documents as actors, so we searched all 12 files for the pattern and rewrote what we found outside quotes, 134 passages in all. We gave the verb to the producer or to ourselves, and left a program as the actor where a program acts. A local script compared the text against 69430330 and found every quoted phrase unchanged. It also reported the signs above, £160m and £954m in place of 160 and 954, new references to Table 3.2 and #55, and one fewer mention of lines 61 and 62. Other number tokens moved only in their punctuation.

Changes in answer to the review of a8f6c299

GPT-6.1 Sol approved a8f6c299 with nine low findings on actors and sentence length, and we took its wording for eight. On the OBR's Table 3.1 we wrote "In two memo lines the OBR also splits", since we name the table at the start of that bullet. We searched the 12 files for documents, keys and fields that give, hold, list, name or carry something, and missed the cases we list in the next section. We compared the text against a8f6c299 and found every number, code span and quoted phrase unchanged, apart from the new reference to the heading "Keys we proposed".

Changes in answer to the review of 207dfecc

An Opus 5.5 cloud session requested changes at 207dfecc, because we claimed in this description to have found no other documents, keys or fields acting, and its search found more. Outside quotes, we still wrote "the repository holds no primary bytes" in 11 notes, and in 19 other sentences we gave "give", "hold", "list", "name" or "carry" to a document, row, figure, field or quoted phrase. We rewrote them all and took the reviewer's wording for most, changing it where a document or row still acted: IFS's addendum and first-lane sentences, TPA's chart group and the OBR's component sentence.

We then searched the 12 files again, outside quotes and code spans, for those verbs in their other forms and for 27 other verbs with a document, row, figure or field as subject. We rewrote what we found, for example "conditions.horizon naming both years" as "we name both years in conditions.horizon", changing 87 lines in all at 0e444f7f. We left people, bodies and programs acting, and arithmetic such as "three times £93.28 gives £279.84".

We also compared 122 row references in the notes with the staged rows using a local script, read the 15 it flagged, and corrected three. We now point to OBR rows 2675 and 2679 for the £0.4bn and £0.5bn on lines 70 and 71, and to red book row 359 for the 60%, where we had said 2673, 2677 and 365. A local script compared the text against 207dfecc and found every quoted phrase unchanged. It also reported those row numbers, new references to paragraphs 3.2 and 3.5 and to component in the OBR notes, and number tokens that moved only in their punctuation.

Changes in answer to the review of 0e444f7f

The same cloud session requested changes, because in 0e444f7f we narrowed IFS's sentence on dropped parameters to a measure's name, while we still keep "£100 Household Energy Payment (Jan 2027)" in the rows' reform_hint. We restored the fact with the reviewer's wording, and took its wording for TPA's line 11 and the OBR's "15 measures expected to reduce the tax gap". In IFS's notes Miller now gives no number "in her remarks", we state README rule 7 as the reason for TW_0's manifest line, and IFS records the dates in the PDF metadata. A local script compared the text against 0e444f7f and reported every quoted phrase kept, those two new quoted phrases, the £100, and one number token that moved only in its punctuation.

The review of e78e1921

The same cloud session approved e78e1921 with two low findings. In the first, a reader could take our IFS sentence "though the coordinator assigned the 29 October briefing" to mean a briefing rather than the deck. We merge e78e1921 unchanged and will use the reviewer's wording, "assigned it as the 29 October briefing deck", in the follow-up PR where we stage the OBR's paragraph 4.30 and Table 6.3 rows. In the second, the reviewer counted two number tokens whose punctuation alone we changed, IFS's "7." for rule 7 and TPA's "11," for line 11, where our script reported one.

What the notes now say about #247's ingest

We replaced each sentence that left a choice to the ingest with what #247's ingest does:

  • It puts Table 6.2's 60 rows on cash_requirement_change, beside OBR row 2684.
  • It drops red book rows 339–350 as twins of uk_hmt_tables rows 264–275. It drops as restatements red book row 185, RF's six fuel duty freeze rows, RF's rows 385–390 and TPA's four restated totals.
  • It merges RF rows 26, 27, 60, 132 and 201 into another printing of each figure.
  • It keeps OBR row 2740 on decisions_effect_on_borrowing, scores TPA's five remaining totals on that metric, and copies derived_from onto the 155 claims that TPA's 186 derived rows become.
  • It scores CenTax's £520m and £640m on the childcare option and rows 58, 108 and 109 on the CGT option, drops rows 19 and 20, and scores HM Treasury's six Exchequer rows on the British Business Bank key.
  • It scores IFS row 46, the £80 from January 2027's Household Energy Payment, on average_household_income_change per household, and tallies rows 43–45 as partial-year support.

A local script, which we didn't commit, traced each staged row through stage() at a28223f2, and we report that run's outcomes above. We also gave today's row numbers in uk_tpa's at-ingest sentence so readers wouldn't mistake kept row 27 for a drop.

🤖 Generated with Claude Code

Keeps every fact; names who fetched, staged and left out each item.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
@vercel

vercel Bot commented Nov 5, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
policyengine-scorecard Ready Ready Preview Nov 5, 2026 7:52pm UTC

Request Review

Keeps every fact; names who fetched, staged and left out each item.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
Keeps every fact; names who fetched, staged and left out each item.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
Keeps every fact; names who fetched, staged and left out each item.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
Keeps every fact; names who fetched, staged and left out each item.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
Keeps every fact; names who fetched, staged and left out each item.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
Keeps every fact; names who fetched, staged and left out each item.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
Keeps every fact; names who fetched, staged and left out each item.
The page-by-page appendix of unstaged numbers stays as it was.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
Keeps every fact; names who fetched, staged and left out each item.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
Keeps every fact; names who fetched, staged and left out each item.
Restores the spaces missing in run-together words ("since2010-11",
"Paragraph3.16", "at2075"), as review B finding 16 asked. The two
key tables stay as they were, apart from those spaces.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
Keeps every fact; names who fetched, staged and left out each item.
The six tables stay as they were.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
Keeps every fact; names who fetched, staged and left out each item.
Restores the spaces missing in the Wernham and Boileau sections
("has18", "OBR3.8%", "seed230,10 of21"). The tables and code
blocks stay as they were, apart from those spaces.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
uk_devolved_govts, uk_hmt_redbook, uk_ifs_ab2026, uk_obr_tables and
uk_rf_ab2026 each kept one inanimate "holds"/"carry" or passive clause.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
Six families had renamed the heading; uk_ippr and uk_tpa's review
notes refer to "the access recipe", and the other six families keep it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6HDMCMjxRvqg56vuU9
@MaxGhenis

Copy link
Copy Markdown
Contributor Author

Ready for hub review

Head: 794635caecc5fa00a1ede7f37f4a45bdbf3d888b (on uk/ab2026-notes-voice, stacked on uk/ab2026-harvest at 92f364c, which hasn't moved).

The diff touches only the 12 NOTES.md files under sources/harvest-uk-ab2026-2026-10-29/, with one commit per family and three small follow-ups. I didn't touch any claims file, manifest, COUNTS.json or README. I kept tables and code blocks as they were, apart from restoring missing spaces in uk_obr_tables and uk_ifs_ab2026. I kept the "Access recipe" and "Not staged" headings because the review notes cite them. The bold "Working files." openers are now plain sentences.

Fact check, per family

I ran a local script, not committed, that pulls numbers, backticked spans, fenced blocks, quoted phrases, URLs and ab2026 keys from each file's old text (origin/uk/ab2026-harvest) and new text, and prints the set differences.

Family Result Words old → new
uk_centax no fact dropped or appeared 2783 → 3013
uk_devolved_govts no fact dropped or appeared 2581 → 2771
uk_hmt_costings no fact dropped or appeared 6342 → 6636
uk_hmt_redbook no fact dropped or appeared 7908 → 8358
uk_hmt_tables no fact dropped or appeared 1270 → 1429
uk_ifs_ab2026 differences, all from restored spaces (below) 9076 → 9795
uk_ippr no fact dropped or appeared 1373 → 1477
uk_obr_tables differences, all from restored spaces (below) 7461 → 8006
uk_rf_ab2026 one case change (below) 7159 → 7525
uk_thinktank_misc no fact dropped or appeared 2409 → 2616
uk_tpa no fact dropped or appeared 3929 → 4170
uk_trussell_wpi no fact dropped or appeared 497 → 551

Reasons for each difference:

  • uk_ifs_ab2026. Quote around35% became around 35%. Numbers dropped: -30%, 108, 140, 470, 5,7,8,14 and 6,7,8,10,11,14,15,18,19. Numbers that appeared: 30%, 0.13, 0.91, 1% and 1.9%. Each difference comes from restoring a space in the Wernham and Boileau sections:

    • -30% subgroup was a list item missing its space, now - 30% subgroup.
    • current£108, printed£140 and EDF£470 now read £108, £140 and £470, so the script tokenises them with the £.
    • pp4,5,7,8,14 and rows1,6,7,… now read as spaced lists.
    • to0.13, and0.91, printed1% and and1.9% now tokenise.

    No value changed.

  • uk_obr_tables. Numbers dropped: 11.7 and 26,49. Numbers that appeared: -11.7, 2075, 3.48, 3.52, 49 and £12. These also come from restored spaces, which review B finding 16 asked for: prints-11.7bn became prints -11.7bn, lines 26,49 and 60 became 26, 49 and 60, and at2075, Paragraph3.48, Paragraph3.52 and to£12bn gained their spaces.

  • uk_rf_ab2026. Quote By became by. The original wrote "By", not "on" about the lead of the landing-page bullet. My rewrite of that lead uses a lower-case "by", so the explanation now quotes "by".

Every proposed key survives whole in its family's file. The tests below include that check.

Tests

  • PYTHONPATH=. uvx --with pytest-xdist --with pydantic --with pyyaml --with hypothesis pytest -n 4 tests/test_uk_ab2026_harvest.py tests/test_ab2026_harvest_contract.py: 433 passed. The default interpreter has no pytest, so I ran it through uvx with the test imports.
  • uvx ruff format --check pipeline tests: 170 files already formatted.

Sentences I left ambiguous, or changed in a way you should check

  1. uk_centax, in the parents_bunching bullet: "(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)". It's unclear whether the metric name copies the adapter's label. I kept the wording.
  2. uk_rf_ab2026, in the Period convention: "period 2028 (the £4.4bn: 2027, as printed or read)". It's unclear which £4.4bn rows were "printed" and which "read". I kept the wording.
  3. uk_hmt_costings, in the p.41 table cell and bullet: "very probably counts this costing inside a Table 6.1 line". I kept "very", because it grades a probability and isn't an intensifier I could cut without weakening the claim.
  4. uk_obr_tables, in the proposed-key table, …state_pension_triple_lock_adjustment: "Figures 3.51 are historical simulations 1990–2025". It probably means paragraph 3.51's figures. I kept the wording and only restored the space.
  5. uk_obr_tables: "Table 3.1 prints-11.7bn" became "prints -11.7bn". I read the hyphen as a minus sign, not a separator.
  6. uk_obr_tables: "Design … 57 rows (58 until …)" and Review fixes item 1, "58 rows dropped", both stay as they were. They count before and after the 4 November restore.
  7. uk_devolved_govts, 3 November finding 2: the original said "(superseded on 4 November, above: …)", but the 4 November section comes after it. I wrote "see the 4 November changes" instead of "above".
  8. uk_ifs_ab2026: the first lane's section counts (Wernham 21 rows, Boileau 28, Adam 18) differ from today's 15, 21 and 12. The file's opening paragraph says the sections above win, and I kept the counts, framed as what "the first lane staged".
  9. uk_thinktank_misc, 29 October check: "all six primary hashes" against today's 5 primaries. On 29 October the family still included Trussell's primary. I kept the wording.

I flagged no other sentence as ambiguous. The rewrite names an actor throughout: we, the lane, the coordinator, a reviewer or the producer. Tables and row lists keep their original form.


Generated by Claude Code

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
At b227956 all 186 derived rows are own, 155 become claims and 31 drop on other grounds.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6Hz6VzBiZKpP6s4opP
At b227956 rows 1, 25 and 99 become claims on the option key; the ingest makes no reference to #207's file.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6Hz6VzBiZKpP6s4opP
At b227956 merge_republications folds rows 26, 27, 60, 132 and 201 into
another printing of the same figure.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LwQq6Hz6VzBiZKpP6s4opP
@MaxGhenis

Copy link
Copy Markdown
Contributor Author

A, B and C applied

Head: 2883dcb6a044cb09043496ab6a0c971691087b30 on uk/ab2026-notes-voice, 13 commits on dbfd766d. git diff --name-only dbfd766d HEAD lists 8 NOTES.md files and nothing else. Tests: pytest tests/test_uk_ab2026_harvest.py tests/test_ab2026_harvest_contract.py -q -n 4 gives 550 passed.

#247 heads I read. I read scorecard_db/ingest_uk_ab2026.py, tests/test_uk_ab2026_ingest.py's test list, KEY_DECISIONS.md and #247's description at b227956. While I worked, #247 moved to 002d90c, which adds 24 lines to tests/test_uk_ab2026_ingest.py and changes nothing else. Neither head changes any claims_staged.jsonl.gz since 3812ce4d.

How I checked the ingest. To state what the ingest does rather than what it should do, I ran #247's stage() at b227956 in a scratch worktree (not committed). A wrapper recorded each staged row's outcome: claim and metric, or drop reason. The run gives read 5,952, ingested 4,686, dropped 1,266, which matches ingest_uk_ab2026_expected.json. Every C sentence states an outcome from that run, plus the code that produces it. One limit: #247's description says build_db.py doesn't call this ingest yet, so the sentences say what the ingest does, not what the site shows.

A. Wernham's dropped metrics (uk_ifs_ab2026)

  • Rows checked. Wernham's 15 staged rows (39–53, here and on Ingest the Autumn Budget 2026 harvest (tranche 2 of #230) #247) carry average_household_support 4, cpi_inflation_rate 2, affected_share 2, and one each of average_energy_bill_change, revenue_change, share_of_energy_price_increase_offset, share_gaining, share_losing, average_annual_gain and real_income_growth. None of the six dropped metrics appears, and no row mentions the bus fare cap. That gives 21 − 6 = 15.
  • Applied as you wrote. The list's opening line, the fare_cap bullet and the "apart from the £2 bus fare parameter" cut.
  • The other five dropped metrics. Each bullet now takes the first lane as actor in the past tense ("the first lane proposed it for …", "the first lane staged … as context").
  • Two more present-tense sentences in the section:
    • The opening paragraph's "We give announced statutory parameters administrative_data …" now reads "The first lane gave …". No kept Wernham row is a parameter.
    • "We treat the 30% subgroup labels, the 5–15 bus-travel age eligibility, … as conditions or context" now ends "…, and no row we stage carries the 5–15 bus-travel age eligibility." Only rows 48 and 52 carry a 30% subgroup condition.
  • Left as it stands. The verification paragraph's sample and its "The 21 component rows pass row_problems", as you asked.

B. Boileau

"We stage them under the proposed metrics … and none under forecast_spending_change.", as you wrote.

C. Tranche 2 statements

Note Now says What it rests on (b227956)
uk_hmt_tables L54–57, heading now "## At ingest" We put Table 6.2's 60 rows on cash_requirement_change, the metric of OBR row 2684. We drop red book rows 339–350 (Table 6.3's repeat of lines 45 and 46) as twins of rows 264–275 and record each under publication.also_printed_in. DISPOSITIONS_AB2026; _twin_key and stage(); outcomes of all 60 and 12 rows
uk_obr_tables L216, head now "At ingest." Table 6.2's 60 rows land on row 2684's cash_requirement_change. same
uk_obr_tables L243 "We don't stage it yet. We stage HM Treasury's Table 6.2 total in uk_hmt_tables." The rest of L239–243 stands. —
uk_rf_ab2026 L428 and the key table's …fuel_duty_continued_freeze cell We drop RF rows 154 and 397–401 as the OBR's forecast restated, and move OBR row 2740 to revenue_change under README rule 10. RF rows 385–390 drop as restatements. ROW_DROPS, ROW_METRICS and outcomes
uk_tpa L195 We copy derived_from onto the 155 claims the 186 rows become, and drop 31 (30 out of scope, 1 ratio). No code in #247 reads the field, so separating them from model estimates stays open. Run: 155 of TPA's 189 claims carry the field. A grep of scorecard_db, app/src and pipeline finds no reader.
uk_tpa L196 We drop the restated rows 3, 27, 147 and 237, and score 26, 169, 4, 28 and 265 on decisions_effect_on_borrowing. The ingest merges none of the five, so the duplication stays open. Outcomes (current rows 3, 26, 146, 236 dropped; 25, 168, 4, 27, 264 ingested); the sentence keeps the notes' pre-drop numbering
uk_hmt_redbook L198 #247 registers the breathing-space key partial with the three readings open (KEY_DECISIONS.md), and we drop row 185 as a restatement. KEY_DECISIONS.md row; row 185's outcome
uk_hmt_redbook L411 (RB-21) Past tense for the 4 November handover, then the twin drop of rows 339–350. same as uk_hmt_tables
uk_centax L145 We drop row 19 (62%) as a ratio and row 20 (88%) as CenTax's earlier estimate restated. ROW_DROPS
uk_centax L147 We drop row 20 and score 58, 108 and 109 on the CGT option key, with scenario on each claim. Run: 3 claims carry the 2024 scenario
uk_hmt_costings L180 and L203 We score rows 179–184 on the BBB key alone and sum no row into a Table 6.1 line or total. Outcomes; stage() sums nothing. Rows 176–178 (PSND/PSNFL) drop as out of scope.
uk_obr_tables L45 Unchanged. It names "the package keys we proposed and tranche 2 registered" and gives a standing rule ("never sum totals and components"), not a tranche 2 task. —

Three more forward-looking sentences of the same kind sat outside your list. I changed them on the same evidence, and you may want to check this call:

I left the general guards "an ingest never adds them to the total" (uk_centax, uk_devolved_govts) and RF's history sentence about the AB2025 ingest ruling.

Token check against dbfd766d

Family What changed
All families with a C edit "2" falls out where "tranche 2" went
uk_hmt_tables 60, 12, 2684, 339, 350, 264, 275 and publication.also_printed_in
uk_obr_tables 60, 6.2, 2684 and cash_requirement_change
uk_rf_ab2026 2740 (twice), 26, 27, 60, 132, 201, and metric spans
uk_tpa the row numbers and counts above (186, 155, 31, 30); derived_from once fewer (L69); the quoted pointer "Review fixes, 3 November 2026"
uk_centax rows 1, 19, 20, 25, 58, 99, 108, 109, "62%", "88%", "#247" and two key spans
uk_hmt_redbook partial, KEY_DECISIONS.md, uk_hmt_tables, publication.also_printed_in
uk_hmt_costings 179 and 184 (twice each)
uk_ifs_ab2026 the six metric names as backticked spans, "3 November" and "rule 8"

Fenced blocks, URLs and hashes match dbfd766d in every file.

Sentence lengths on what I changed

The 62 added sentences have mean 20.2 words, median 20 and maximum 40, with none over 40. The be-verb and "be + above/below" greps on the added text find nothing.

Open, as the notes now say


Generated by Claude Code

A PR or a tranche doesn't act, so "we" now takes each of these verbs.
No key, count or row number changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The sentence kept the numbering from before finding 1's drop, so row 27 read as
dropped when today's row 27 holds the kept £19.1bn claim.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
MaxGhenis and others added 2 commits November 5, 2026 08:55
…ayment

Give the ingest's own reason for moving OBR row 2740 to revenue_change: the
Government didn't adopt the freeze. Point the OBR notes at row 2740, not 2737,
for the continued freeze's effect on borrowing. State that the ingest tallies
IFS's £140, £24 and £36 and scores the £80 on average_household_income_change.

Name HM Treasury, the OBR, RF, IFS or ourselves where a row, table, page or
document acted, in the sentences the review listed and across the 12 files.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@MaxGhenis
MaxGhenis changed the base branch from uk/ab2026-harvest to main November 5, 2026 14:08
…ctors

- Restore the five minus signs the OBR prints on the savings and property
  income tax rows, and split the sentences over 40 words.
- State that OBR row 2740 stays on the borrowing metric, as #247 now does.
- Search all 12 notes for documents, tables, rows, pages and keys acting as
  subjects, and give the verb to the producer or to us.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ences

Give the verb to the producer or to us in the nine sentences GPT-6.1 Sol
named, and split one wrapper around a 36-word quotation. Every number and
code span stays as it was.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The cloud review of 207dfec found the description's search claim false:
outside quotes, 11 notes still said "the repository holds no primary bytes",
and 19 other sentences gave one of five verbs to a document, row, figure, field
or quoted phrase. We rewrite those sentences, the same verbs in their other
forms, and 27 other verbs with such a subject, and give each verb to the
person, body or program that acts.

We also correct three stale row numbers. We now point to OBR rows 2675 and 2679
for the £0.4bn and £0.5bn on lines 70 and 71, and to red book row 359 for the
60%, where we had said 2673, 2677 and 365.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
We restore IFS's sentence on dropped parameters, which we narrowed to a
measure's name in 0e444f7, and take the reviewer's wording for TPA's line 11
and the OBR's "15 measures expected to reduce the tax gap". We also name who
acts in three more IFS sentences.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@MaxGhenis
MaxGhenis marked this pull request as ready for review November 5, 2026 21:15
@MaxGhenis
MaxGhenis merged commit 3557c2b into main Nov 5, 2026
15 checks passed
@MaxGhenis

Copy link
Copy Markdown
Contributor Author

We merged e78e1921 with a merge commit, under the merge-your-own-PR gates.

  • Independent review. An Opus 5.5 cloud session approved e78e1921 with two low findings (hub-notes/pr246-review-e78e1921). We answered the count in the description, and we'll use its IFS wording in the follow-up PR where we stage the OBR's paragraph 4.30 and Table 6.3 rows. The same session requested changes at 0e444f7f, and we answered that review in e78e1921. Subfleet held the Codex lanes this afternoon, so no GPT-6.1 Sol review ran at this head.
  • CI. gh pr checks returned 0 at this head, with all 15 checks listed, and GitHub reported the merge state as clean.
  • Scope. In this PR we changed the notes for all 12 families. We changed no staged row, no code, no file under app/, and no site or lane data.

Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com

@MaxGhenis
MaxGhenis deleted the uk/ab2026-notes-voice branch November 5, 2026 21:15
MaxGhenis added a commit that referenced this pull request Nov 6, 2026
Apply #246's L1 wording and preserve the IFS notes corrections from main. Refresh the OBR key summaries and scope for the seven new rows so both families' notes match main byte for byte.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

This branch was successfully deployed

1 active deployment
Preview — e78e1921 Deployed Nov 5, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants