[juaristi22]

Summary

microcosm.data.load("uk") now follows latest-national.json (#823). That pointer names the certified national cut promoted on 26 October (microcosm-uk-2024-25-national-20261025T091414Z-a6299504, built on policyengine-uk 2.151.0). The default used to be the June 2023 populace_uk_2023 artifact, which stays at load("uk", 2023). The national variant key retires.

This changes the default UK load for every consumer of microcosm.data. It moves out of #1422 so that it can land on its own (Vahid's review of #1422), and it depends on nothing there.

• The registry entry's variant becomes the default variant, so resolve("uk") and load("uk", 2025) follow the line's pointer.
• The loader, registry and release-assembler tests and the release runbook follow. The live pointer test now reads the line from the pointer, not from the variant.
• The unreleased changelog entry from Per-line release pointers, the UK 2024-25 national default and its promotion command (#823) #966, which described the opposite state, is removed, and so is the last sentence of 823-uk-2025-registry.changed.md, which said the default stays on the June 2023 line (Vahid's review of UK: take up #1095's actionable items on policyengine-uk 2.165.1 #1422).

Tests

• On this branch (main at 98aef7529 plus 003dbd006; 45df16363 changes only a changelog fragment), the 18 test files that resolve or load a UK dataset or name the default variant: 1,246 tests pass.

🤖 Generated with Claude Code

[vahid-ahmadi]

Review of 45df163 (draft, so a comment). Thanks for splitting this out of #1422; it does what my #1422 review asked.

No musts or shoulds.

Checked

• registry.py: the 2025 entry moves to DEFAULT_VARIANT, so latest_year("uk") is 2025 and resolve("uk") follows latest-national.json. The June 2023 artifact stays at ("uk", 2023) on latest.json, and the tests pin both.
• The line still comes from the pointer path, not the variant (_line_from_pointer_path), so the release assembler's default_datasets: {"national": ...} and the promotion runbook are unaffected.
• The two stale changelog texts from UK: take up #1095's actionable items on policyengine-uk 2.165.1 #1422 (the Per-line release pointers, the UK 2024-25 national default and its promotion command (#823) #966 fragment and the last sentence of 823-uk-2025-registry) are gone, and the runbook now describes the flip as done.
• No other PolicyEngine repo calls variant="national" (org code search; it finds only this repo's files). The one outside caller of microcosm.data.load (bottom-50-tax-analysis) loads US data, so nothing downstream breaks.
• CI is green on every job at this head.

Nits

1. The changelog entry says the promoted cut was built on policyengine-uk 2.151.0. It could add one clause that a default load on today's lock (2.165.1) therefore lacks the inputs added since, for example the ISA subscriptions (Budget 2026 ISA data, PR B of #1418: ISA subscriptions per adult, the Lifetime ISA anchor, additive ISA income, policyengine-uk 2.164.2 #1421) and company cars (UK: company cars from the FRS job table (dashboard #316) #1430), until the next cut is promoted. People reading "default load" may assume it matches the lock.
2. Retiring the national key is a removal. If it ever shipped in a release, a .removed.md fragment would say so; if it never did (the body says the Per-line release pointers, the UK 2024-25 national default and its promotion command (#823) #966 entry was unreleased), ignore this.

Verdict: ready to merge once it leaves draft.

What I ran: read the full diff; shallow clone at the head to grep for remaining national variant uses and the loader's variant handling; GitHub code search across the org. I relied on CI for the tests and didn't run them locally.

🤖 Generated with Claude Code