WS-C deferred adjudications and the reconciliation layer

[juaristi22]

Central registry of the adjudications deferred during the #622 build-out (PR #735) and the calibration-target deferrals accumulated since, to be reviewed as the spine (#145/WS-E) and calibration workstreams converge on the first certified cut. Each item links its signed rationale in the committed resources (uk/target_reference_membership.json, uk/ledger_compile_parity_*_signed_differences.json, uk/calibration_measure_exclusions.json). Items graduate to their own issues when they become actionable. Spine-lane counterpart: #796.

Deferred adjudications from the #622 build-out

• Property-income ×1.9 undercount → reconciliation home. The incumbent scales SPI Table 3.7 amounts ×1.9 (trace: policyengine-uk-data#311 / policyengine-uk-data#230 — SPI covers only taxpayers with liability; HMRC Property Rental Income Statistics show £46.68bn vs SPI ~£24.5bn for 2020-21). The family is signed-excluded from activation. Doctrine-clean paths: a chronicle package for the rental statistics + a declared reconciliation, or a declared value-scaling operation citing Add coverage-universe targets for uninsured and private coverage #230. This is the anchor case for the reconciliation layer.

• Region-grain activation for the 18 two-level contract targets — Fixture A's 131 region×age production rows are structurally unmatched while activation pins country grain.

• ons.population.scotland_households_3plus_children — the contract wants a household-composition fact that does not exist at ≤2025; needs a chronicle-side fact or an adjudicated alternative source.

• TCL year surface — the incumbent pins the April-2025 publication values at {2026}; the Ledger side activates at 2025. Revisit when a 2026-vintage publication exists.

• CGT 2025 level → UK CGT: translate the 2024-25 gains outturn to the 2025 calibration year (forestalling year; OBR CGT +44.2% signed on the stack register) #875 (2026-09-05). Superseded facts: since the Chronicle re-pin (microcosm#874) the latest outturn is 2024-25 (£127.3bn of gains, 584k taxpayers — the forestalling year ahead of the October 2024 rate rise) and the doctrine binds it at face value at the 2025 period; the engine's 2025-26 rates on those gains overshoot OBR's FY2025-26 receipts by +44.2% (obr.capital_gains_tax@2025). Ruled option 1: facts stay bound as published, the OBR miss is signed on the publication-stack register (obr.capital_gains_tax@2025, adjudication UK CGT: translate the 2024-25 gains outturn to the 2025 calibration year (forestalling year; OBR CGT +44.2% signed on the stack register) #875, window 2026-09-05 → 2026-10-05). Deliverable in UK CGT: translate the 2024-25 gains outturn to the 2025 calibration year (forestalling year; OBR CGT +44.2% signed on the stack register) #875: the index ruling for a declared 2024-25 → 2025 translation (rate-adjusted; a derived Chronicle fact), applying uprating_index in the compile step, the taxpayer count flat at 584k, and the cash-lag mapping for self-assessed OBR lines.

• The 347 uprating holds — currently explicit identity holds; per-family index declarations belong to the materialization layer.

• Salary-sacrifice canonical source — HMRC-table facts vs the incumbent's OBR-forecast rows (value ratios 0.81–1.14); pick per target when calibration binds them. (The five measure exclusions covering the family are windowed to 2026-11-25.)

• UC payment-band source-only rows — 12 'No payment' / 'or over' facts skipped in fan-out (no safe incumbent inverse).

• UC family-split month choice — +2.6–15% drift from latest-month resolution vs the incumbent's pinned publication month; candidate for calendar-average treatment as fact history grows (the caseload total already moved to the calendar-year average per the UK scorecard: populace_uk_2023 vs enhanced_frs_2024_25 vs admin benchmarks (2026) #731 ruling; fold uk-data#466's family-type/caseload sum-consistency in here).

• Calibrate the UK national build from Ledger-backed targets #623 disposition — resolved: the calibration seam landed via Add the UK national calibration step over ledger-backed target references #729/UK national calibration run-readiness: doctrine, live parity trio, US-format diagnostics, scorer, staging posture (#623) #743, its gate story completed in PR Complete the #757 deferrals: the exclusion pass, the measured U8 lever, the release-cut certification, and the re-armed thresholds #793, and Calibrate the UK national build from Ledger-backed targets #623 closes with that PR (residual acceptance work tracked here and in WS-E spine deferrals: known bugs and signed inconsistencies awaiting their increment #796).

• TCL retirement at the year move — the limit is abolished from 2026-04-06 (Budget 2025; policyengine-uk-data#465), so at period 2026 the 15 TCL targets are structurally zero: the references must retire or re-vintage with the calibration-year move (WS-F/Retarget the UK build to FRS 2024-25: re-pin raw vintages, regenerate parity instruments, re-measure gate baselines #723). Add the contract-side guard from uk-data#465's step 2: a test refusing a reference whose binding variable is structurally zero at the reference period. (Our 2025 activation is unaffected — the limit is in force in FY 2025-26; when uk-data#465 moves the incumbent's values to 2025, Fixture B's 15 TCL ledger-only additions become matched rows at the next pin refresh.)

• UC element-count targets (housing, LCWRA, carer, childcare — policyengine-uk-data#466, Re-measure UC composition errors after first calibrated run; element references still unsourced #738 step 3): no contract declarations exist; pipeline is chronicle facts first (childcare's published 160k series needs no API key — easiest first package; the rest need Stat-Xplore UC_Households exports), then a contract amendment per the Refresh the UK Universal Credit target contract #727 precedent, then references auto-activate on regeneration. Facts requested: UC element-count and monthly composition facts for the microcosm UK target surface chronicle#187.

• water_and_sewerage_charges raw-mapping divergence — resolved by fix-and-sign during the Assemble the microcosm-built UK spine: whole-spine parity and swap acceptance #686 lane: verified on the raw FRS 2024-25 tab (12,644 vs 14,307 nonzero households, all 1,663 differing households Scottish), the defect is the incumbent's NaN-propagating addition of the retired CSEWAMT header, and the candidate side stands. Signed as scottish-water-incumbent-nan-zeroing (defect_fix, +0.1009) on uk/spine_swap_signed_differences.json with a quantitative bound; latent in the incumbent since at least 2023-24 and surviving at 1.56.16.

• Census household counts bind at the 2021/22 vintage in the first rowwise local candidate (revision 2026-09-02, microcosm#762 A11). The Declare cross-grain target reconciliation for the UK local surface: national wins, sub-national rescales (#495 increment 6 companion) #802 bridge that reconciles the OA ladder's census household counts (650 constituencies + 361 local authorities) to the national household-composition partition cannot bind while three of its ten cells (multi_family_households, unrelated_adult_households, lone_parent_non_dependent_children_households) are UK spine: relationship-to-head frame column, so the ONS household-composition split is measurable #791's reviewed measure exclusions (in force to 2026-11-26). The rowwise candidate therefore classifies that bridge as unbound with a receipt (cross_grain.unbound_bridges in the candidate plan and manifest) and binds the 1,011 census household rows as published — 2021 for England, Wales and Northern Ireland, 2022 for Scotland — while every national row binds at the 2025 calibration period. The vintage gap is a known level pull (−2.58% household mass measured at K=2 in July); it is re-measured in the First calibrated rowwise UK local candidate (#495 increment 6) #762 receipts. Resolution rides UK spine: relationship-to-head frame column, so the ONS household-composition split is measurable #791 (the relationship-to-head column) or a shares/uprating ruling on the local family; until then the deferral is deliberate and visible, not silent. Revised 2026-09-03 (microcosm#762 A15): the shares/uprating ruling landed — every ladder household row now binds uprated to the calibration year by one national factor from the Ledger's ons.households_total fact (29,003,000 for 2025 over the ladder's 28,060,832 = 1.033577), receipted in the candidate manifest and mandatory under --release-candidate; the bridge stays unbound until UK spine: relationship-to-head frame column, so the ONS household-composition split is measurable #791. See the next item.

• Five reviewed measure exclusions for national rows the spine cannot reach by reweighting (microcosm#762 A16, ruling 2026-09-03; window 2026-09-03 → 2026-10-03; erratum in the 2026-09-03 comment below). Register uk/calibration_measure_exclusions.json (49 entries): ons.savings_interest_income (spine £13.8bn vs ONS D.41 £95.5bn, HMRC SPI £18.3bn — a frame gap plus a concept gap) → UK spine: savings interest income a third under HMRC SPI and five times under the ONS D.41 row — reconcile the target concept and review FRS interest capture #866; obr.housing_benefit (£2.7bn vs £12.1bn) → UK spine: housing benefit £2.7bn against OBR £12.1bn — pensioner and supported-accommodation HB missing after the UC migration #867; slc.borrowers.plan_2_liable and slc.borrowers.plan_2_above_threshold (−65 % / −69 %) → UK spine: SLC plan-2 borrower stocks at a third of SLC's — the FRS flag captures repayers, not the liable stock #868; dwp.jsa_claimants (2.8×) → UK spine: JSA claimants 2.8× the DWP caseload — legacy-benefit reporting after the UC migration #869. The two ONS land rows first listed (8.7× / 2.5×) were withdrawn the same day: their misses were an artefact of per-block engine resolution in the candidate driver (receipt R15), and single-block resolution puts them within 3 % and 9 % of the ONS balance sheet. Each exclusion fails the run on expiry: correct the spine (the issues) or re-adjudicate. Also ruled the same day and applied on uk-rowwise-candidate-762-b: stretch bound 10, grain_equal, K=15 and 1500 epochs as UK local doctrine constants; A17, the census tenure cells uprate with the ladder rows.

• Local target surface: 68 signed deferrals covering 513 cells (uk/local_target_reference_membership.json → signed_deferrals; no expiry — each retires when the fact lands or the frame supports it). Of 21,519 candidate references, 19,105 are active; 314 have no fact at or before the period, 1,586 no fact for the area, 513 are signed. By reason: msoa_mean_to_la_deferred (3 targets, 1,083 area-cells: ONS equivalised income is MSOA-grain means, LA aggregation waits on the mean-aggregation design — Port UK local target surface and credibility gates from uk-data #147 lineage, no issue of its own); council_tax_voa_scotland_absent (256 cells: the VOA feed has no Scottish band counts — a chronicle-side fact ask, no issue yet); council_tax_wales_country_control_absent (176 cells, First calibrated rowwise UK local candidate (#495 increment 6) #762 A13: no Wales country stock-by-band fact — asked for in the 2026-09-03 comment, no issue yet); council_tax_ni_domestic_rates (88: NI has domestic rates, structurally absent); private_rent_pipr_* (4 reasons, 361 cells: the PIPR feed is 2026-06 only, after the period, missing four English authorities, Scottish BRMA grain, no NI — chronicle-side, no issue yet); uc_gb_only_ni_absent / uc_children_gb_only_ni_absent (101 cells: Stat-Xplore UC local facts cover GB only — no issue yet); spi_la_target_measure_coverage_absent / spi_pcon_self_employment_mean_absent (9 cells: SPI publishes medians only for E06000027, nothing for Scilly, one constituency lacks the SE mean — structurally absent); council_tax_city_of_london_band_a_suppressed (1: VOA suppression); local_authority_support_floor_excluded (43) and council_tax_band_h_spine_support_absent (296) are spine-side and tracked on WS-E spine deferrals: known bugs and signed inconsistencies awaiting their increment #796.

• Local binding adjudications (uk/local_binding_adjudications.json, 6 entries; Declare cross-grain target reconciliation for the UK local surface: national wins, sub-national rescales (#495 increment 6 companion) #802 2026-08-31 → 2026-11-30, First calibrated rowwise UK local candidate (#495 increment 6) #762 A1 2026-09-02 → 2026-12-02): the fences lifted so the local families may bind — census disclosure-control noise; the SPI frame-model proxy; full-FRS TEI bands unavailable; population universe = private households; UC bound at benefit-unit grain; VOA dwellings vs the household frame. Each is a target-side reconciliation the family stands on; renewal or a real fix falls due at expiry. No issue yet.

• Local uprating holds (17,077 cells; uprating_holds in the local membership): age 2024, SPI tax-year 2023 and the census tenure cells 2021/22 are held to 2025 as identity except tenure, which A17 uprates with the ladder rows; UC 2025-05 and council-tax stock 2025 carry none. Same resolution as the national 347 holds above: per-family index declarations in the materialization layer, or later facts (asked for in the 2026-09-03 comment).

The uk_target_fit disposition remainder (PR #793's adjudication queue)

The adjudicated exclusion pass (42 windowed entries, expiring 2026-11-26, correct-or-renew at apply time) landed in PR #793; the first certified cut blocks on what remains:

• The 8 UC caseload/two-child-limit cells (−28.6%…−62.0% at the post-exclusion optimum): the pre-registered U8 take-up lever is measured and exhausted (0.85 leaves every cell unchanged while destroying housing benefit −0.0% → −52.4%); the measured lever is capital-test support — UK WAS stage: split private pension wealth out of corporate_wealth (uk-data#452 M2) #750 / policyengine-uk#1824. Decide: wait for UK WAS stage: split private pension wealth out of corporate_wealth (uk-data#452 M2) #750, retarget, or exclude.
• hmrc/dividend_income_count_income_band_500_000_to_1_000_000 (−36.7%): its amount sibling carries a sparse-band exclusion; the count regressed once the sibling unbound. Candidate 7th sparse-class exclusion, or ride the Band edges derive from the exclusion-pruned register, silently widening surviving banded targets #792 fix.
• Band edges derive from the exclusion-pruned register, silently widening surviving banded targets #792 — measure materialization is exclusion-set-dependent: 12 of 361 shared measures move (up to +858% initial) when the register roster changes; 349 bit-stable. Pick the fix direction (per-variable RNG isolation vs materialize-full-register-always) — blocks certifying on the surface.
• Doctrine posture for the certified cut: the Phase-3 run carries epochs 1500 / family_equal as receipted overrides (the 2026-08-24 ruling keeps uniform the default); the certification records them verbatim. Promote to doctrine, or certify with receipted overrides.
• Fuel-duties ledger retarget (household-incidence fact vs total receipts — chronicle-side): the binding is verified correct; the gap is universe scope plus LCFS diary under-capture at 37% of implied litres. The target carries a windowed exclusion meanwhile.
• The publication-stack target-fit exclusions fall due 2026-09-30 (uk/target_fit_reviewed_exclusions.json on uk-publication-stack-835, PR UK spine: benefit-unit-grain SPI UC reporter draw with pre-take-up screen (#832); exact capital sentinel (#833) #835; five entries tracked on WS-E spine deferrals: known bugs and signed inconsistencies awaiting their increment #796): dwp.uc.households_single_with_children, dwp.uc.households_children_1, dwp.uc.households_children_2, dwp.uc.households_children_5_or_more (the UC with-children support residual — eligible support 4.24m benefit units vs 6.76m DWP; the honest-grain reporter of UK spine: condition SPI UC receipt on benefit-unit composition and add a UC-specific child count #832 stands, the draw-shape follow-up is UK spine: composition-faithful SPI income draw — bridge predictors and rank-preserving replacement (with-children UC residual) #840) and hmrc/private_pension_income_count_income_band_100_000_to_150_000 (solver-pressure drift, +25.5 % at the v13 seam). Target-side: renew with a re-measure or fix via UK spine: composition-faithful SPI income draw — bridge predictors and rank-preserving replacement (with-children UC residual) #840 before the window closes; this is the first UK window to expire. Sixth entry (2026-09-05, on uk-publication-stack-834, microcosm#874 composition): obr.capital_gains_tax@2025 (+44.2%, the 2024-25 forestalling-year gains bound at 2025 — UK CGT: translate the 2024-25 gains outturn to the 2025 calibration year (forestalling year; OBR CGT +44.2% signed on the stack register) #875), window 2026-09-05 → 2026-10-05. The Port the post-freeze childcare / Tax-Free Childcare inputs and targets, re-pin the Chronicle feed, and activate the DfT bus targets (#834, #789) #874 run-4 receipt also reports the private-pension band entry stale (+24.7%, back inside the bound; the rot rule wants it removed) and two unsigned state-pension SPI bands (50–70k +34.6%, 20–30k +25.3%) outside the bound.
• Exclusion-window renewals fall due 2026-11-25/26: the 5 salary-sacrifice and 42 target-fit entries all fail closed on expiry — each needs its gap corrected or its adjudication renewed by then.

National calibration-target issues previously untracked here

• Adjudicate the incumbent's four post-calibration aggregate scalers #790 — adjudicate the incumbent's four post-calibration aggregate scalers (the rail/bus/fuel class deliberately left out of E6 scope)
• UK bus fare and subsidy consumer targets from Chronicle facts #789 — UK bus fare and subsidy consumer targets from Chronicle facts
• Re-measure UC composition errors after first calibrated run; element references still unsourced #738 — re-measure UC composition errors after the first calibrated run; element references still unsourced
• Certified UK artifact carries less than half of admin UC caseload (3.26m vs ~6.7m; £36.9bn vs £75.8bn) #701 — certified UK artifact carries less than half of admin UC caseload (the predecessor measurement behind uk-data#452/UK WAS stage: split private pension wealth out of corporate_wealth (uk-data#452 M2) #750)
• UK rebuild: sequence calibration after the amounts imputations; staging seam has no CGT target parity evidence #704 — sequence calibration after the amounts imputations; staging seam has no CGT target parity evidence
• UK scorecard: populace_uk_2023 vs enhanced_frs_2024_25 vs admin benchmarks (2026) #731 — the scorecard (populace_uk_2023 vs enhanced_frs_2024_25 vs admin benchmarks) — becomes the acceptance probe of the first certified cut
• Publish Chronicle package IDs in calibration diagnostics targets #661 — publish Chronicle package IDs in calibration diagnostics targets

Refs: #622, #735, #665, #145, #796, PR #793.

[juaristi22]

Item 10 resolved: #623 narrowed to run-readiness (not closed) — the tooling is delivered by draft PR #743 (doctrine, live parity trio, US-format diagnostics with the chronicle pin, #578 rule-1 scorer, staging-candidate posture, runbook); the licensed armed run is held until WS-E E8 (#684) + E10 (#686) complete or María names a different base. #623 stays open for the run + acceptance evidence; the aggregate-scaler ruling (item context) remains open on #623 itself.

[juaristi22]

Registry coverage: 29 contract targets have no reference row — verified findings + adjudication queue

Folding the #623-lane registry findings here per María's ruling (2026-08-25): the coverage work does not ride PR #743, which ships as calibration-seam-only. Everything below was verified mechanically against the pinned feed (chronicle-uk-artifact-1cab809, facts 226358e7…, 107,550 rows), the real compiler (compile_ledger_target_references), the real generator (tools/generate_uk_target_references.py), and policyengine-uk's variable registry.

The gap, exactly

uk_national_targets.json declares 189 contract targets; target_references.json carries 388 references covering only 160 of them. The other 29 compile into no calibration, no score, and no coverage receipt. verify_register_coverage.py reports 388/388 because it counts references, not contract targets — a true statement about the wrong denominator.

Nuance worth keeping: the gap was never invisible. The membership report records all 29 as non-active candidates (417 = 388 active + 27 no_fact_at_or_before_period + 1 multi_fact + 1 signed_excluded). It was recorded but unenforced — nothing turns an unexplained coverage hole into a failure.

Root causes (verified by re-running the generator against the current feed: it reproduces 388/29 exactly)

1. 27 of 29: the generator's geography-pin heuristic. _geography_id_for_target infers a pin from target-id substrings and merges it over the contract selector, so the selector matches zero facts:
   • obr.council_tax_{england,scotland,wales} → pinned E92/S92/W92 while OBR files everything at K02000001 (the split is by concept, not geography); obr.domestic_rates → pinned N92000002 by substring rule, facts at K02000001.
   • slc.* → pinned GB (K03000001) vs the facts' England (E92000001) for borrowers/support and S92/W92/N92 for the devolved repayments.
   • dwp.pip.* → pinned GB vs the facts' England+Wales (K04000001).
   • scotgov.* → falls to UK ("scottish" ≠ "scotland" as a substring) vs S92000003.
2. Multi-year SLC/SCP concepts additionally need period pins: chronicle's layouts embed the year, so eligible facts across years carry distinct semantic keys and the resolver's latest-match refuses. The selector vocabulary already supports period_value; with pins, everything compiles (verified 42/42 in a synthetic run).
3. The 9 Scottish council-tax-stock selectors also lack source_concept (feed carries scotgov.chargeable_dwellings_band_a…_h/_total, Scotland, month 2025-09) — both fixes needed.

What is wireable — and what is not

Wireable now: 29 references closing 27 targets (verified end-to-end, values resolved):

• 4 OBR (council tax E £41.9bn / S £3.05bn / W £2.40bn 2025 projections; domestic rates £494m),
• 4 SLC borrowers (plan-2 liable 9.71m / above-threshold 4.46m; plan-5 230k / 35k — 2025 projections),
• 6 SLC support (maintenance 1.16m recipients / £8.59bn; ADG 18.6k / £55.4m; PLA 99.6k / £181.4m — 2024 actuals),
• 3 devolved repayments (S £203.3m / W £229.1m / NI £181.7m),
• 2 PIP daily living (standard 1,283,000 / enhanced 1,608,000 — Jan 2025, England+Wales),
• 1 Scottish child payment (£471.0m — 2025 projection, not the £455.8m 2024 actual),
• 9 Scottish council-tax stock bands (band A 498,707 … total 2,623,149).

All 29 bindings' variables verified present in policyengine-uk (866 variables — pip_dl_category, council_tax_less_benefit, student_loan_plan, council_tax_band, etc.).

NOT wireable — the 13 property-income amounts. They sit under the generator's standing signed exclusion: the incumbent's amount target carries the ×1.9 property-income undercount adjustment (policyengine-uk-data #311/#230; HMRC PRIS £46.68bn vs SPI ~£24.5bn), so binding raw SPI facts would knowingly calibrate to ~10/19 of the incumbent's concept. They stay out pending a chronicle PRIS package or a first-class value-scaling operation (chronicle#196 lane).

Blocked — ons.population.scotland_households_3plus_children: selector reaches 30,925 age facts and cannot answer a household question; the right concept (ons.families_households_table7) has no Scotland cut. Filed as chronicle#196; worth neutralising so it fails loudly.

Adjudications needed (María)

1. Property-income signed exclusion — keep signed out (default) or overrule? If kept: route is chronicle PRIS package / value-scaling op.
2. Devolved repayments contract split — the binding note "the resolved geography binds the model country filter" describes the incumbent's pipeline; microcosm's materializer has no geography→filter translation. Proposal: split slc.repayments.devolved_total into three contract targets with explicit country binding filters (the OBR council-tax pattern). Contract count 189→191.
3. PIP geography — facts are England+Wales; the binding counts pip_dl_category model-wide, so Scotland's ADP claimants inflate the estimate against an E+W target. Add a country filter to the two bindings, or record a signed asymmetry?
4. Forecast-period convention — OBR/borrowers/SCP resolve 2025 projections under allow_source_projection (the naive compile silently yields the prior-year observation instead). Confirm projections are the intended targets, matching the existing 24 OBR references.
5. Generator changes — honor contract-declared geography before the substring heuristic (heuristic stays as fallback so the existing 388 regenerate byte-identically), plus period_value selector pins for borrowers ("2025"), slc.support ("2024"), SCP ("2025").

Mechanics when this lands (one commit, after the digest-neutral seam work)

Regenerate — never hand-edit: target_references.json + membership (generator), and the two parity signed-differences resources (tools/build_uk_ledger_compile_parity_signed_differences.py). The parity fixtures describe the June/incumbent sides and do not change; the 2023 gate already tolerates 2025-only references (211/388 compile at 2023 today). Pinned counts that move: ACTIVE_REFERENCE_COUNT 388→417, per-family membership counts, contract target count 189→191. All legacy_json — no spec/gate-battery digest movement.

Plus the regression guard: a test on the committed JSONs (CI has no feed) asserting every contract target has ≥1 reference or a membership status in {signed_excluded, blocked-with-tracking-issue} — turning the already-recorded statuses into an enforced contract.

Re-score both artifacts on the new register before quoting any head-to-head; the v9 figures (297/71/10, losses 0.085 vs 0.235) don't carry across a denominator change. Closes ~20 of the 65 incumbent-only national rows; the 13 property amounts join the fact-gap bucket.

🤖 Posted from the #623 working session (Claude Code).

[juaristi22]

Correction: I measured against a stale base — E2 is already fixed on main

My comment above was computed against the #743 branch, which sits 245 commits behind main. Measured against main the numbers are different, and one of my two work items is already done.

Against main today: 397 references covering 169 of 189 contract targets — 20 unreferenced, not 29.

@MaxGhenis's d9d5a45e ("Pin the scotgov council-tax stock targets to Scotland", 2026-08-23) already wired the nine scotgov.council_tax_stock.* targets, and diagnosed the same root cause I described — the substring geography rule sees no "scotland" in "scotgov", pins to the UK, and the pin is merged into the selector and matched against fact.geography.id, which Chronicle stamps S92000003. Membership moved 388 → 397 active, no_fact_at_or_before_period 27 → 18. So E2 is closed; disregard that section.

Two things survive the correction, and one is worth deciding deliberately.

The remaining 20 break down as: 13 property-income amounts (standing signed exclusion — unchanged), 1 blocked (ons.population.scotland_households_3plus_children, chronicle#196), and 6 groups still unwired: OBR council tax ×3 + domestic rates, SLC borrowers ×4, SLC support ×6, devolved repayments, PIP ×2, Scottish child payment. All still verified wireable against the pinned feed, with the same adjudications outstanding (devolved contract split, PIP England+Wales geography, forecast-period convention).

The fix strategy question is now live rather than hypothetical. d9d5a45e fixed the geography rule with a targeted prefix special-case:

if target_id.startswith(SCOTGOV_COUNCIL_TAX_STOCK_PREFIX):
    return UK_GEOGRAPHY_IDS["scotland"]

That is correct and well-documented for those nine. But each remaining group needs its own special-case under that approach — OBR files at K02000001 while the rule infers E92/S92/W92 from the target name; SLC borrowers are England while the rule infers GB; PIP is England+Wales; the devolved rows are per-country. That is five more branches in a heuristic whose failure mode is silent: a wrong pin yields zero matches, and a zero-match target is deferred rather than raised.

The alternative I verified: let a geography_level/geography_id declared on the contract's own ledger_selector win over the inferred pin, keeping the heuristic as the fallback. The existing references regenerate byte-identically (the heuristic still answers wherever the contract is silent), and each future target states its geography where a reader would look for it. Worth a ruling before anyone writes the sixth special-case.

Everything else in my comment above stands — the compile verification, the property-income exclusion, the regeneration inventory, and the regression guard, which is the item that would have caught this class in the first place.

Apologies for the stale baseline; I should have measured against main rather than the working branch.