Release the 10/8 uk-data batch (15 PRs, relock to policyengine-uk 2.122.2)

[MaxGhenis]

Release the 10/8 uk-data batch as one data release, expected to be 1.58.0. This branch contains 15 member PRs at their reviewed heads, the planned merge conflict resolutions, the SPI UC pension-age integration (I1), and the dependency relock (R). It lands only on Max's go under d833.

The member PRs' impact estimates were measured separately, under different policyengine-uk versions or development heads. They do not add, and no full build of the combined release has run locally.

Member PRs and exact-head reviews

PR	Change	Reviewed head	Independent review
#503	Exclude lodgers' rent payments from property income.	0812cd1b3c5952a2b4b259ae046baf44dfced8a3	approves this head
#509	Exclude property losses from income and count sub-let rent for every tenure; carries #503.	cc937e0555d7ead58fc980b41b1959510709b1db	approves this head
#490	Calibrate Housing Benefit to DWP's GB figures by age group.	f4d8f95c512ad47b902c2a2bb9c12f8853cce7b8	approves this head (delta review over r4 at 954c225)
#510	Solve Pension Credit take-up over entitled units and use 37% take-up for newly entitled units; carries #490.	65656305498b1964ca000ecf1620097eec71c5ec	approves this head
#516	Weight BRMA draws by census private-rented households.	29e1bd7a18a80942cc9014946d2dc8bea7e40b44	approves this head
#513	Use FRS benefit-unit capital for Pension Credit's capital test.	a1690afaa72595b2d5a847ac6581b692be1fbf14	approves this head
#536	Fall back to committed OBR workbooks when obr.uk returns a non-workbook page.	7605e63efb57b1dce40df91ea459aa7da9485cc6	approves this head
#530	Replace OBR UC targets with one GB OBR UC total and reach the UC payment top band in calibration.	974fa8cbb90b636fdc06e21195a420a3e0fad7cf	approves this head
#533	Restore HMRC salary-sacrifice relief targets.	0863a6aa540be49dac65cfd73c17ebe82b17eecd	approves this head
#514	Remove unobserved benefit claims from SPI-synthetic rows.	deb16e944e0ed00df43f6492344ebbc6b055657c	approves this head
#517	Draw each cloned household's Output Area from its own FRS region.	1c5b20c02a2f0d3509fff90cc679f302b8225c0e	approves this head
#524	Supply is_claimant_or_partner from the FRS adult table.	5e320ff2140bf2cb2a7ad0c01bc31d5fa71cb766	approves this head
#525	Set UC gainful self-employment from FRS main-job status.	dc2abf7b4cd52b9e63c3631020af6e7e0c79c6da	approves this head
#526	Map FRS EMPSTATI 11 to OTHER_INACTIVE.	1832adeb2296b53fe5a7d7194e770118aa34d1f8	approves this head
#529	Draw SPI incomes within earnings groups set by FRS employment status.	d39371d508c33065e636fd9dcc5f15e750177fa2	approves this head

A4 merge conflict resolutions

• Weight FRS BRMA draws by census private-rented households #516: union both .gitignore sides, retaining one local_geography_weights.csv.gz line and .brma-cache/; reset uv.lock to main's b45c373c version.
• Use the FRS benefit-unit capital for Pension Credit's capital test #513: resolve the one pyproject.toml hunk with policyengine-uk>=2.107.0, then set the final floor in R; retain the clean auto-merged lock until R.
• Calibrate GB universal credit to one OBR total; reach the UC payment top band #530: retain Calibrate Housing Benefit to DWP's GB figures by age group #490's targets/schema.py hunk with carry_forward; Calibrate GB universal credit to one OBR total; reach the UC payment top band #530's conflicting hunk repeats the older countries version.
• Stop SPI-synthetic rows carrying benefit claims nobody observed #514: retain assign_reported_takeup(...), Calibrate Housing Benefit to DWP's GB figures by age group #490/Solve Pension Credit take-up over entitled benefit units #510's comments, and Calibrate Housing Benefit to DWP's GB figures by age group #490's & ~derive_all_claimants_over_state_pension_age(...) exclusion on the base-FRS would_claim_uc draw; union .gitignore and reset the lock to main's.
• Supply is_claimant_or_partner from the FRS adult table #524: retain both new functions in datasets/frs.py and reset the lock to main's.
• Set UC gainful self-employment from the FRS main-job status #525: retain all three functions in datasets/frs.py, including derive_uc_is_in_gainful_self_employment.
• Map FRS EMPSTATI 11 to OTHER_INACTIVE, not LONG_TERM_DISABLED #526: retain Calibrate Housing Benefit to DWP's GB figures by age group #490's person_benunit_id, is_adult, and is_SP_age fixture branches and Map FRS EMPSTATI 11 to OTHER_INACTIVE, not LONG_TERM_DISABLED #526's [66] * len(...); remove the obsolete lha_list_of_rents.csv.gz fake reintroduced by Map FRS EMPSTATI 11 to OTHER_INACTIVE, not LONG_TERM_DISABLED #526, as Weight FRS BRMA draws by census private-rented households #516 deletes that file.
• Draw SPI incomes within earnings groups set by FRS employment status #529: reset its conflicting uv.lock to main's before R.

#503, #509, #490, #510, #536, #533, and #517 merged without conflicts. No code choices beyond the A4 rules were needed.

I1: SPI donor UC pension-age exclusion

Apply #490's pension-age exclusion to #514's SPI donor redraw, so a benefit unit whose claimants are all over State Pension age cannot acquire would_claim_uc through that redraw.

apply_spi_donor_benefit_rules(dataset, donor_person=None, uc_pension_age_excluded=None) accepts an optional boolean array with one entry per benefit unit. After redrawing take-up flags, it applies would_claim_uc &= ~uc_pension_age_excluded when the array is supplied; with None, it excludes nothing. The rule function does not construct a Microsimulation.

The sole pipeline caller always computes the mask on the real target dataset with the existing _all_claimants_over_state_pension_age(target_dataset, year) helper, using the same computation as create_frs, and passes it explicitly. The helper loads a one-year UKMultiYearDataset container holding a copy of the target, so it evaluates that year's claimant and State Pension age variables without extending the dataset or uprating unrelated household inputs. Example and property tests compute the mask on _people_dataset(...) and pass it in; a caller-wiring test captures the keyword argument so omitting it fails. This keeps synthetic fixtures independent of household uprating inputs and avoids adding a simulation to every donor-rule Hypothesis example.

#490's exclusion therefore reaches both UC draws. #510's enhanced-dataset Pension Credit take-up step still supersedes #514's SPI Pension Credit redraw. On policyengine-uk 2.122.2, claimant_or_partner_variable selects is_claimant_or_partner, supplied by #524.

R: dependency relock

Set the dependency floors to policyengine-uk>=2.122.0 and policyengine-core>=3.32.13, retain one hypothesis>=6.168.3 requirement in dev, and keep build and towncrier>=24.8.0 on separate lines. Lock policyengine-uk exactly to 2.122.2, the version covered by the pin analysis, rather than advancing to a later release. #525 requires policyengine-uk 2.122.0 or later.

The expected lock changes are exactly four package records:

Package	Main's lock	Release lock
policyengine-uk	2.93.0	2.122.2
policyengine-core	3.31.1	3.32.13
hypothesis	absent	6.168.5
policyengine-uk-data	1.56.16	1.57.4

R adds changelog.d/uk-data-batch-relock.changed.md. Together with the 15 member fragments, versioning should produce one 1.58.0 section containing 16 entries; #524's .added fragment drives the minor bump from 1.57.4.

Local validation

The four targeted files passed: 48 passed, 1 skipped (39 warnings) in 120.46 seconds, using the required shared lockf -k, two workers, and a fresh temporary directory removed afterward. The pension-age example, property, and caller-wiring tests passed; the original synthetic-fixture regression passed unchanged. The dataset-dependent pension-age check skipped because the built dataset was unavailable.

uv lock --check passed for 162 packages. Ruff formatting passed for 242 files and git diff --check passed. All 15 reviewed heads are included; only the seven expected storage paths differ, with no generated geography output.

What publishes after the go

The release branch push publishes no data. The release PR runs PR CI on the combined tree. On Max's go, one merge to main starts versioning and a docs redeploy; versioning commits 1.58.0, and push.yaml runs the full build and tests. The docs redeploy runs independently and can complete even if the later build tests fail. The hub cancels the duplicate push.yaml run on the merge commit at the old version.

If the full build and tests pass:

• Private Hugging Face model repo policyengine/policyengine-uk-data-private: one commit writes seven root files: base FRS, enhanced FRS, tiny-base FRS, and tiny-enhanced FRS .h5 files; parliamentary_constituency_weights.h5; local_authority_weights.h5; and local_geography_weights.csv.gz. It also writes release_manifest.json at the root and at releases/1.58.0/, then creates and verifies HF tag 1.58.0.
• GCS bucket policyengine-uk-data-private: overwrite those same seven objects in place, each with version metadata 1.58.0. Consumers fetching the HF root files or GCS objects by name receive the new release.
• GitHub: main receives the release merge and the Update package version commit with version 1.58.0 and one CHANGELOG section. The publish workflow attempts a lightweight git tag 1.58.0, which the hub verifies afterward. The gh-pages docs redeploy comes from the merge commit because Weight FRS BRMA draws by census private-rented households #516 and Draw each cloned household's Output Area from its own FRS region #517 change docs/.

HF uploads before GCS. The installed-version manifest certifies policyengine-uk==2.122.2 and policyengine-core==3.32.13. Nothing is uploaded to PyPI or to the public HF repo, and no GitHub Release object is created. This batch changes no release-path files. The downstream microcosm#1087, PE-UK #2082, and policyengine.py release follow on their own gates; d778's policyengine.py release pins the then-latest policyengine-uk and core with the new data and certifies that bundle.

The release PR carries the member PRs' issue closures because GitHub's handling of their closing keywords when the member heads reach main has not been verified:

Fixes #515
Fixes #504

🤖 Generated with Claude Code