[juaristi22]

Summary

The local-area cut microcosm-uk-2024-25-dense-20261107T173210Z-38605444 was built at 03c31df56, the head of this branch. The branch is two commits on top of dc00d158f, the merge of #1535. PolicyEngine/policyengine.py#744 certifies that cut on policyengine-uk 2.174.1, and its review asked for these two commits on a reviewed PR before it leaves draft.

• edd051ecc defers one local council tax cell, Carmarthenshire's band I, where the cut's first build stopped.
• 03c31df56 signs dense build 1's gate-only target-fit failures for the local line.

Neither commit changes the national line. The deferral touches local references only, and every sign-off is scoped to release_lines: ["local"].

Please merge with a merge commit, not a squash. Then 03c31df56, the cut's recorded build commit, stays reachable from main after the branch is deleted.

edd051ecc: defer Carmarthenshire's band I council tax cell

• Why the first build stopped: the local build on dc00d158f stopped at uk.full.problem. Carmarthenshire (W06000010) has a nonzero Welsh band I council tax target, but the K=15 dense pool holds no band I household there, so the rowwise driver's support check refuses the cell.
• The deferral: tools/generate_uk_local_target_references.py defers welshgov.council_tax_stock.by_area.band_i for W06000010. It uses reason council_tax_band_i_spine_support_absent and defer_if_compiles=True, as Shetland's band H was deferred under UK council tax target family: re-base the stock rows on the councils' taxbase returns (chronicle#262 / #264), bind Wales, Scotland and band I, compose the region controls, and track the band-support and level gaps #929.
  • María ruled it on 7 Nov. UK: every open data item: uk-data parity, engine locks, the local surface, consumption and high-value homes #1095 records the ruling (comment), with a check to retest the cell when the spine or K changes.
  • The deferral lives in the generator, so the next local cut from main keeps it.
• Regenerated on the local cut of feed 64753803:
  • local_target_references.json, with the cell's reference removed;
  • local_target_reference_membership.json;
  • the local compile-parity receipt, ledger_compile_parity_local_incumbent_2025_signed_differences.json.
• Tests: test_uk_council_tax_targets.py and test_uk_battery_bindings.py follow the deferral.

03c31df56: sign dense build 1's gate-only failures for the local line

• Build 1: uk-local-candidate-f100-s42-20261107T102622Z-f52640d9, at edd051ecc with review date 2026-11-07. It stopped at the terminal gates on uk_local_target_fit alone.
• The register changes, drafted by tools/draft_uk_dense_signoff.py and approved by María on 7 Nov:
  • 5 national target-fit rows and 6 local target-fit cells are signed for the local line, each expiring on 5 Dec 2026;
  • 13 stale entries are retired, because their rows came back inside the bound, which the gate fails on;
  • the 6 area-support entries stay.
• What build 2 did: every signed row stays bound in the solve. Build 2 reused build 1's solve (UK local-area release: publish a K=15, 150,000-household dense dataset in at most two builds #1301) and passed all 7 terminal gates and every release-blocking gate. Build 2 is the cut.
• Tests: test_uk_terminal_gates.py follows the registers.

Known limitation

On this cut, four of the national ONS single-year rows between 19 and 22 come out short:

• women aged 19 by 32%, where the national cut eeb6093e is 2.5% short;
• men aged 19 by 22%;
• men aged 21 and 22 by 15% each.

ons.population.female_19 is signed for the local line until 5 Dec, and the others are inside the 25% bound. Constituency and local-authority breakdowns of measures that turn on these ages rest on that short count, for example the National Living Wage's 18 to 20 rate or Universal Credit's under-25 standard allowance. #1095 tracks it (comment).

What merging changes on main

• Local-line builds from main carry the Carmarthenshire deferral, and these sign-offs until 5 Dec.
• National-line builds don't change.
• Not renewed here: the local line's sign-offs that lapse on 16 Nov still lapse then.

Testing

• On 03c31df56: test_uk_battery_bindings.py, test_uk_council_tax_targets.py and test_uk_terminal_gates.py give 170 passed and 1 skipped (JUnit).
• Against main: the branch merges cleanly. Main has two commits since dc00d158f: Calibrate federal student loan balances to Federal Student Aid's portfolio (#1074) #1357 and Say where UK national targets come from in fiscal_targets' docstring #1542.
• The cut itself: build 2 at 03c31df56 passed every release-blocking gate. The cut is published on the inspect lane, tag only.

Disclosure

Counts of target rows and cells only, plus one local council tax cell named with no count or value. No unit-record values, and no fit values for local cells.

🤖 Generated with Claude Code

[vahid-ahmadi]

Review of 03c31df

I checked both commits against the cut's build record on Hugging Face, the branch against main, and the registers at dc00d158f and at head. I also read the published gate report of build 2 (…20261107T173210Z-38605444, read-only, aggregates only below) and compared it with the national cut eeb6093e's calibration diagnostics. I didn't run the tests here; CI covers them. The counts below come from the JSON files directly. Paths are relative to packages/microcosm-build/src/microcosm/build/.

Should

1. ons.population.female_19 is signed on the local line at 32% below ONS, though the national cut fits it to within 3% (uk/target_fit_reviewed_exclusions.json:125).
   • It is the first ONS population row the national register has ever signed (git log -S finds no earlier one). UK November event data, PR1: full-time education, single ages 16 to 24, Scottish PRIS diagnostics, energy kWh columns #1535 added the row when it split the 18–19 band into single years.
   • In build 2's uk_target_fit details, the local cut puts 265,017 women aged 19 against ONS's 389,642 (-32.0%). The national cut eeb6093e has 379,809 (-2.5%), from a starting estimate of 255,863. So the local solve leaves the row about where the national one started.
   • Men aged 19 pass, but only just: -22.1% on the local cut against -0.06% on the national. Men aged 21 and 22 are both about 15% low. Of the 31 local cells signed at head, 19 are age/20_30 cells, and all 19 are low.
   • Failure case: constituency and local-authority results from 38605444 count about 115,000 fewer women aged 19 than the national results from eeb6093e do, and about 125,000 fewer than ONS. Any local breakdown of a policy that turns on age, such as the National Living Wage's 18 to 20 rate or the under-25 UC standard allowance, would rest on the short count. The entry's reason is the generic template, and nothing tracks the gap. The 20_30 cells do have a long trail of earlier sign-offs.
   • Suggested fix: name the shortfall as a known limitation in this body and in Certify the Microcosm UK pair from microcosm#1535 on policyengine-uk 2.174.1 (national eeb6093e, local area 38605444) policyengine.py#744's local section, and give it a line in UK: every open data item: uk-data parity, engine locks, the local surface, consumption and high-value homes #1095 (or its own issue). Whether the next local build changes the row's loss weight or pool support is your call. This PR only needs the gap on record before the 5 Dec expiry.

Nit

2. The Carmarthenshire ruling is cited to UK: every open data item: uk-data parity, engine locks, the local surface, consumption and high-value homes #1095, but UK: every open data item: uk-data parity, engine locks, the local surface, consumption and high-value homes #1095 doesn't record it, and the deferral has no review point (tools/generate_uk_local_target_references.py:666-681, rationale at :677).
   • UK: every open data item: uk-data parity, engine locks, the local surface, consumption and high-value homes #1095's latest comment is from 5 Nov, and nothing on it mentions Carmarthenshire, W06000010 or band I. The body's "María ruled it on 7 Nov (UK: every open data item: uk-data parity, engine locks, the local surface, consumption and high-value homes #1095)" points to nothing.
   • The deferral sits in the generator with defer_if_compiles=True and no expiry, the same as Shetland's band H. But its rationale is specific to one spine ("on the spine of microcosm#1535's merge"). With defer_if_compiles, the stale-deferral check at target_reference_authoring.py:679-687 is skipped, so a cell that compiles stays deferred. AreaSignedDeferral does have an expires_on (target_reference_authoring.py:176), but only the small-cell rule path sets it (tools/generate_uk_local_target_references.py:283), and the generator's add() (:521-532) doesn't take one. Failure case: a later spine that does draw band I households in Carmarthenshire keeps the 72-dwelling cell unbound, and I found no check that would flag it.
   • Suggested fix: post the ruling on UK: every open data item: uk-data parity, engine locks, the local surface, consumption and high-value homes #1095 with a checkbox to retest the cell when the spine or K changes, or pass an expires_on through add() for this deferral.

Checked and fine

• Build commits: the cut's build_manifest.json records code.git_commit 03c31df565276e58… with git_dirty false. That is this head, and its parent is edd051ecc. The PR carries exactly those two commits on top of dc00d158f.
• Main: the branch is 2 ahead of main and 2 behind. The two main commits are Calibrate federal student loan balances to Federal Student Aid's portfolio (#1074) #1357 (US student loans, plus US label additions in microcosm-calibrate) and Say where UK national targets come from in fiscal_targets' docstring #1542 (a docstring in uk_runtime/fiscal_targets.py). Neither changes UK build code or UK data, so a main merge isn't needed for this cut.
• Merge commit: 1affbc127, dd99a1dcd and 5293966d4 from Sign off the local cuts' target-fit rows (dbacb023, 83802786), and the company car base #1491 are all ancestors of main, so that merge kept them. Squash is still allowed on the repo, so the body's merge-commit note still matters.
• Deferral: one reference is removed (welshgov.council_tax_stock.by_area.band_i@W06000010, value 72). Active references go from 27,168 to 27,167, and signed deferrals from 2,931 to 2,932. The compile-parity receipt drops the cell's one ledger-only row (7,890 to 7,889). The Wales-wide band I row stays bound, and build 2 fits it to within 0.8% of 5,342, so Carmarthenshire's 72 dwellings are still counted in the Welsh total. The test changes (27,167, 4,180, 21 Welsh band I cells, 2,510 band cells) match.
• Registers: national goes from 11 to 13 (5 added, 3 retired) and local from 35 to 31 (6 added, 10 retired), so 13 retired in all, matching the body and changelog. Every new entry is juaristi22, local only, approved_on 2026-11-07 and expires_on 2026-12-05. Build 2's gates report 0 failing, 0 stale, 0 expired and 0 premature against 1,545 national and 27,167 local targets. So the 13 retired rows are inside the bound or no longer bound, and leaving them in would have failed build 2 as stale. The manifest's reviewed_exclusions_stood_on lists exactly the 13 and 31 entries at head.
• Signed gaps: the five new national rows sit between 26.3% and 46.5% from target, against the 25% bound. Four are HMRC income-band rows and one is the population row in item 1. The six new local cells are all at or past the bound, and one of them only just.
• Lapsing entries: 4 national and 11 local entries expire on 16 Nov, as the body says.
• Changelog: both fragments are present, and their counts match the registers.

CI at 03c31df: lint, wheels, installed-emitter, select-countries and integration-uk pass. engine-free, engine-uk and engine-us (3.13 and 3.14) are still running.

Verdict: the commits are the ones the cut was built from, and the counts hold. Put the 19-year-old shortfall on record before #744 leaves draft.

🤖 Generated with Claude Code

[juaristi22]

Thanks, Vahid. Both are now on record, with no code change in this PR.

1. The figures match the cut's own diagnostics: women aged 19 at −32.0% (−2.5% on eeb6093e), men aged 19 at −22.1%, and men aged 21 and 22 at −15.1%. This body and Certify the Microcosm UK pair from microcosm#1535 on policyengine-uk 2.174.1 (national eeb6093e, local area 38605444) policyengine.py#744's local section now name the shortfall as a known limitation. UK: every open data item: uk-data parity, engine locks, the local surface, consumption and high-value homes #1095 tracks it (comment), with a check to settle the rows' loss weight or the pool's support before the next local build and the 5 Dec expiry.
2. UK: every open data item: uk-data parity, engine locks, the local surface, consumption and high-value homes #1095 now records the Carmarthenshire ruling of 7 Nov (comment), with a check to retest the cell when the spine or K changes, and the body links it. An expires_on through add() can come with that retest.