[MaxGhenis]

WIP: waits on #208 (this branch is cut from it).

The first comparables records under #208's gate, for PolicyEngine analyses published in September and October 2026 before the gate existed. Each is a backfill search of every listed producer, recorded under sources/analysis-<country>-<slug>/ with the analysis's own figures pinned to their results files.

Records

Record	Published	Claims	Compared	Gaps over the threshold
analysis-uk-energy-autumn-budget	2026-10-14	30	9 compared, 21 beside an estimate	4, all explained
analysis-uk-free-childcare-reform	2026-09-03	17	0 compared, 12 beside an estimate	none
analysis-uk-marriage (household calculator)	2026-09-11	97	0 compared, 0 beside an estimate	none
analysis-uk-middle-east-war-living-standards	2026-09-25	34	2 compared, 3 beside an estimate	none
analysis-uk-triple-lock	2026-10-16	35	16 compared, 16 beside an estimate	4: 3 explained, 1 pending
analysis-us-2025-supplemental-poverty-measure	2026-09-18	26	11 compared, 15 beside an estimate	none
analysis-us-child-poverty-impact-dashboard	2026-09-15	13	1 compared, 1 beside an estimate	none
analysis-us-futa-wage-base-dashboard	2026-09-17	51	1 compared, 40 beside an estimate	none

Calls a reviewer should check

• Triple lock, DWP's £15bn. The dashboard is 26% below DWP for 2039-40. The search explained most of it by the size of the survey's pension bill, from measured bills and one paired run, but the analysis's own issue, DWP's £15bn vs our £8.4bn: real paths near £15bn, and two untested explanations (OBR wedge, pension bill) uk-triple-lock#13, says the cause is not established. The gap is therefore pending on that issue: the triple lock dashboard may not be promoted (newsletter, social, press) until Baseline as a first-class attribute of every score #13 is closed and the cause written.
• IFS's £4bn to £20bn for 2039-40 is compared as a range with the expected saving (£11.14bn, inside it), reading IFS's figures as cash, as the record already reads IFS's £4bn for 2034-35. IFS does not say.
• Bases and price bases. Where a backfill job did not record whether a producer's figure allows for behaviour, the claim's basis is unstated and it is not compared. Where it compared a claim without recording a basis (the energy and Middle East VAT and levy costs), the claim is recorded as static, the dashboard's basis, with a note saying so.
• Salvaged record. The 2025 SPM post's search job was cancelled after it wrote its record; the record is kept as written and brought to the current rules.
• Marriage calculator. A household calculator with no fixed estimate (calculator); 97 claims staged for later analyses, none compared, 57 waiting on vocabulary decisions (a null metric with a proposal).

How the records were brought to the current rules

The six records pushed on 21 October, and the salvaged one, predate #208's second round. Each was brought to the current validator by a reproducible script that starts from the pushed record (kept outside the repository with the session's notes): one-word bases with the original text in basis_note, programme slugs with the original text in program_note, results_pins, range groups, one-sided bounds, the DWP real-terms table as prices: real, RF's earnings-link figures as their own world, and percent normalisations. No staged value, quote or source was changed, apart from the Garfinkel paper's source slug and publication date (2022-09-23, from Crossref) and the IFS Green Budget's date (2026-10-12, as the Middle East record dates the same PDF).

Checks

• Every record passes comparables check; the index is rebuilt from them.
• Full pytest tests/ on Python 3.12 with six of the records: 2,491 passed, 15 skipped. With all eight on Record every comparable external estimate before an analysis is published #208's current head: test_comparables, test_export_populations, test_harvest_ingest and test_build_chain, 233 passed.
• The database rebuilds from scratch; populations, lanes and the app copies are regenerated, and the pinned export counts rise by the records' 258 rows (168 UK, 90 US).
• ruff format --check: clean.

Still to come in this PR: the searches for the 2025 SPM prediction post and the targeted energy discount dashboard. The US list has no Census Bureau entry yet; the SPM record found Census's anchored-threshold paper by open search, and adding Census to producers.json is a follow-up.

🤖 Generated with Claude Code

[vercel]

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

Project	Deployment	Actions	Updated
policyengine-scorecard	Ready	Preview	Oct 22, 2026 6:05pm UTC