UK source packages, wave 2: national fiscal/benefit families + uk_national target profile

[juaristi22]

Lanes: packages under ledger-source-ingestor; the profile PR under ledger-target-profile-author (judges: ledger-target-profile + ledger-boundary)
Depends on: #132 (parity fixture established); #131 for the slc_repayments package; Stat-Xplore extracts follow the existing api_response artifact pattern — no new policy needed

Context

policyengine-uk-data (pinned ref for this migration: ebf733c, 2026-07-26) compiles its national calibration registry from 24 source modules under policyengine_uk_data/targets/sources/. Wave 1 ports the three families production last calibrated against; this issue ports the remainder of the national registry, one publication table per package, and then declares the uk_national target profile over them.

Ruling on vintages (2026-08-05): every source-backed original vintage is eventually portable — porting a fact never activates it; Populace owns the active subset. Tranche 1 = current vintages below. Historical vintages (earlier EFO rounds, prior SPI editions, …) land later through a standing backfill lane; the vintage-suffixed package-ID convention (obr/efo_march_2026_*) guarantees no collisions. Uprated/aged/PolicyEngine-computed values never enter Ledger (docs/adr-ledger-facts-only.md).

Package inventory (uk-data module → package)

uk-data module (lines / Target() ctors)	package(s)	notes
obr.py (730 / 13, ~28 realized targets × FY2024–30)	obr/efo_march_2026_receipts, obr/efo_march_2026_expenditure	assertion: source_projection, period_type: fiscal_year. uk-data commits the EFO workbooks under storage/obr_efo/ because obr.uk 403s CI runners — R2 archival is the proper fix for the same problem; ingest from the archived workbooks.
dwp.py (323 / 21)	dwp/benefit_stats_feb_2026, dwp/pip_stats_jan_2026, dwp/benefit_cap_nov_2025, dwp/two_child_limit_apr_2025, dwp/uc_payment_distribution	Values are hand-transcribed in uk-data with reference URLs — the resolve_pending.py first-print lane fits most of these. UC payment distribution comes via Stat-Xplore: capture the API response as the artifact (sheet_name: api_response precedent, 5 existing packages; Census-CD API precedent for the extraction_method wording). A free DWP Stat-Xplore key is needed only by whoever runs the extract.
hmrc_spi.py (366 / 4)	hmrc/spi_collated_2324	Same collated ODS populace's national build parses natively (hmrc_income.py); one package serves both eventually, but do not touch populace's HMRC stage in this issue.
hmrc_cgt.py (268 / 4)	hmrc/cgt_statistics_2024	Single-homing: populace's native UK_CGT_TARGET_SPECS (populace/build/uk_runtime/fiscal_targets.py) and uk-data's module both retire onto this one package (populace retirement happens in PolicyEngine/populace#622).
hmrc_salary_sacrifice.py (129 / 3)	hmrc/salary_sacrifice_tables_6_1_6_2	Publisher CSV already.
slc_repayments.py (100 / 1)	slc/loan_repayments (or extend the wave-1 SLC package)	Academic-year periods — needs #131.
voa_council_tax.py (300 / 4)	voa/council_tax_stock_2025	National band counts; LA-level stock goes to wave 3.
ons_households.py, ons_tenure.py, housing.py (275 / 5)	ons/families_households_2024, ons/tenure_2024	Curated extracts; the ONS families & households datasets also lack stable URLs — same lane as SNPP.
ons_savings.py (61 / 1)	ons/savings_interest_haxv	ONS time series (HAXV, UKEA).
ons_public_sector_employment.py (51 / 1)	ons/public_sector_employment_2026	New module since June.
nts_vehicles.py (49 / 3)	dft/nts_vehicle_ownership_2024	Publisher is DfT (National Travel Survey); new agency directory.
scottish_government.py (37 / 1)	scottish_government/budget_2026_27
mhclg_regional_land.py, ons_land_values.py, _land.py (169 / 4)	ons/land_values, mhclg/regional_land	Verify the actual publisher/table for each before packaging — _land.py is a shared helper and the land series have moved between ONS/MHCLG publications.

Survey-derived aggregates (any APS/LCF-based table encountered) additionally name survey_instrument per the README provenance rules.

The uk_national profile

After the packages above land: one PR in the ledger-target-profile-author lane adding policyengine_ledger/target_profiles/uk_national.json —

• selectors referencing the raw facts above (deterministic check: selector references raw Ledger facts),
• measurement + bindings.policyengine mirroring uk-data's targets/compute/* semantics (entity, map_to, filters),
• no values (FORBIDDEN_VALUE_KEYS in target_profiles/model.py), no runtime hooks,
• counterfactual/cross-tab targets (two-child limit, UC family-type splits, benefit cap) get declarative binding payloads only — the metric providers live in populace, following the JCT neutralize_variable precedent in populace's us/fiscal_target_references.json.

Acceptance

• Every package: manifest source_url + sha256 + R2 URI; cell-geometry guards; concept_evidence_url per measure; deterministic checks green; judge lanes as headed.
• The uk_national profile validates with no embedded values and resolves every selector against built facts.
• A consumer artifact covering waves 1+2 resolves every national-registry row that uk-data compiles at ebf733c for the current calibration year, with any intentional differences (vintage updates, dropped rows) enumerated in the PR description — no silent gaps. (The mechanical compile-parity gate itself lands populace-side.)

Out of scope

• Local-geography families (wave 3). Historical-vintage backfill (standing lane, separate issues as needed). Populace consumption/retirement (Consume Ledger UK facts: uk_target_references, mapping module, and a compile-parity gate populace#622, Calibrate the UK national build from Ledger-backed targets populace#623). The legacy db/etl_obr.py / db/etl_ons.py dicts — superseded by these packages; their cleanup is tracked in UK enablement tooling: source-inventory de-hardcode, UK consumer-artifact suite, legacy ETL cleanup #135.

Sizing

~13–14 packages, batched per agency ≈ 5–6 PRs + 1 profile PR.