Replay the five events on policyengine.py 6.2.6 and diff against the legacy run

[MaxGhenis]

Stacked on #163, which carries #160–#162. W6 target: the five events rerun on the latest certified pair, with a diff report, by 19 October.

What this is

The five OBR fiscal-event replays (Spring Budget 2023 to Spring Statement 2025) rerun on certified variant pe6.2.6_uk2.125.1_policyengine-uk-data-1.58.0:

• policyengine.py 6.2.6;
• policyengine-uk 2.125.1 and core 3.33.0;
• enhanced FRS 2024–25 from uk-data 1.58.0 (SHA-256 5ba399ea…60b6).

The run went through the qualification path, pipeline/uk_replay_qualify.py, with a released_bundle request for the 6.2.6 PyPI wheels (stored as REQUEST.json).

• Execution. The environment was the variant's exact freeze. The probe matched the committed certified record. Wall time was about 18 minutes with two workers on the shared Mac.
• Coverage. 85 measure-year artifacts, and every supported grid is complete. CY2023 falls outside the population's data years (2024–2030), so AS2023, SB2024 and SB2023 lose their FY2023–24 counterparts. Those rows stay visible as unsupported_on_variant.
• Context files. CONSTRUCTION_PREMISES.json and the re-verified MODEL_DIAGNOSTICS.json sit beside the events.
• Seal. NAMESPACE_MANIFEST.json binds every file in the namespace and the diff. REPLAY_RECEIPT.json has receipt_sha256 4feed765…f9afa and answers request_id 6b65b744…eb20. CI verifies the seal, regenerates every comparison and re-renders the diff byte for byte.

Release diff: legacy (pe 5.0.2 / uk 2.89.2 / populace-uk-2023) → 6.2.6

results/uk/replays/diffs/<legacy>__<6.2.6>/DIFF.md lists everything, and packets/ has 78 investigation packets.

Flag	Rows
numerical (> max(£1m, 1%))	304
structural	15
construction	6
support	34
coverage, mapping, inertness	0

This diff changes the engine and the data family at once. No move is attributable to either alone. Certified variants are exact pairs, so the runner refuses to cross them. Future candidates from release qualification usually change one side, and their diffs will attribute.

Measure totals summed over the common years 2024 onward, in £bn, gain to the Exchequer. These are descriptive, not a verdict.

Event	Measure	Years	2.89.2 / populace	2.125.1 / eFRS 1.58.0
AB2024	CGT main rates	2024–29	+24.79	+12.52
AB2024	Employer NICs package	2024–29	+82.06	+86.59
AB2024	Private-school VAT	2024–29	+6.62	+10.40
AB2024	SDLT surcharge +2pp	2024–29	+10.92	+11.74
AB2024	Winter Fuel means test	2024–29	+3.70	+2.56
AS2023	Class 1 employee 2p cut	2024–28	−59.35	−57.58
AS2023	Class 2 abolition	2024–28	−3.91	−2.04
AS2023	Class 4 1p cut	2024–28	−2.91	−1.55
AS2023	LHA reset	2024–28	−3.94	−9.63
SB2023	Annual Allowance package	2024–27	−22.19	−12.56
SB2023	UC childcare caps	2024–27	−0.01	−0.92
SB2024	Class 1 employee 2pp cut	2024–28	−59.35	−57.58
SB2024	Class 4 2pp cut	2024–28	−5.82	−3.10
SB2024	HICBC threshold and taper	2024–28	−8.62	−7.30
SS2025	UC health element	2024–29	+1.67	+4.09
SS2025	UC standard allowance	2024–29	−1.65	−2.95

Structural and construction flags:

• AS2023 Class 2 abolition. The Class 4 head falls to zero (−£786.8m → 0 in CY2027). The legacy diagnosis traced that head to a strict-comparison cap formula on 2.89.2. Whether 2.125.1 fixed the formula or moved the base is for the packet's investigating lane.
• SB2024 HICBC. Child Benefit gains a counterpart (0 → −£273m in CY2024). This is consistent with the opted-out claims fix that the legacy diagnosis recorded as landing after 2.89.2; the packet says how to confirm it.
• AB2024 CGT. The annualisation premise changed (2.125.1 day-weights the rates), so the frozen reversal leaves the blended FY2024–25 increase in both worlds. A revision is proposed in docs/uk_replay/CONSTRUCTION_REVISIONS.md; it waits on cos d1344.

Packets go to the coordinating session, which decides what to file upstream. This pipeline files no issues.

axiom: n/a: replay results, no policy encoding.

🤖 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 11, 2026 8:20pm UTC