Pin England-scoped SLC and scotgov targets to their publication geographies

[MaxGhenis]

Follow-up to #754. That PR fixed the nine scotgov.council_tax_stock.* geography pins and deliberately left the other 14 no_fact_at_or_before_period holds unadjudicated. This PR adjudicates all 14 individually and activates the 11 that are mechanical pin bugs, leaving 7 deliberate holds with their rationales recorded below and in the pin-rule comments.

Per-id adjudication

Every hold was diagnosed against the pinned consumer feed (consumer_facts_uk.jsonl, sha 226358e7…, chronicle at 1cab809) by matching each target's non-geography selector and listing the geographies, periods and assertions of the facts it reaches, then checking the contract binding and the policyengine-uk model code for the country scope.

Fixed here (11):

target id(s)	facts in feed	why the pin was wrong
scotgov.scottish_child_payment_spending	S92000003, fy2024 outturn + fy2025/fy2026 budget projections	Same bug class as #754: the nation-substring rule sees no "scotland" in "scotgov"/"scottish_child_payment" and fell through to the UK pin. The benefit and the facts are Scotland-scoped. Resolves to the fy2025 projection (£471m) under allow_source_projection.
slc.borrowers.plan_{2,5}_{above_threshold,liable} (4)	E92000001, academic years 2024–2029 (projections)	The contract bindings already filter country == ENGLAND explicitly, and the facts are England-stamped; the GB default pin could never match. Resolve at AY2025.
slc.support.{maintenance_loan,parents_learning_allowance,adult_dependants_grant}_{recipients,spend} (6)	E92000001, academic years 2013–2024 (observations)	The bound model variables are England-gated by construction (maintenance_loan_in_england_system, parents_learning_allowance_eligible, adult_dependants_grant_eligible all require country == ENGLAND in policyengine-uk), and the facts are England-stamped. Resolve at AY2024 with uprating holds 2024→2025 recorded.

Deliberately still held (7): these are not pin bugs and activating them via a pin change would be wrong.

target id(s)	why held
obr.council_tax_{england,scotland,wales}, obr.domestic_rates	Facts exist at K02000001 (OBR Table 4.1 nation rows, chronicle#192) versus nation pins, but the joint verdict from the council-tax review stands: keep held until the UK council-tax binding is redesigned (the active obr.council_tax target is row 19, which includes NI domestic rates and an accruals adjustment with no household counterpart; the redesign should prefer publisher administrative targets or row 15).
dwp.pip.daily_living_{standard,enhanced}_claimants	The FOI fact covers England and Wales (K04000001), but the binding has no country filter and the model's pip_dl_category is derived from FRS-reported amounts with no country restriction, so the model metric's footprint is GB. Activating with a K04000001 pin would compare a GB metric to an E&W count. Needs a binding decision (an England-and-Wales country filter), not a pin change.
slc.repayments.devolved_total	Wrong twice over: "devolved" means Scotland + Wales + Northern Ireland (not GB, the current pin), the feed carries one fact per nation (no aggregate), and the bound metric name slc/student_loan_repayment/devolved is not computable by the compute path (its country map has no "devolved" entry). Needs a per-nation redesign of the contract target, not a pin.

Changes

1. tools/generate_uk_target_references.py — replaces the Pin the scotgov council-tax stock targets to Scotland (activates the nine held CTAXBASE references) #754 stock-prefix special case with a TARGET_PREFIX_GEOGRAPHY_PINS table: scotgov. → Scotland, slc.borrowers. → England, slc.support. → England, each entry commented with the layer that fixes the geography (fact stamp, contract binding, or model country gate). slc.repayments.* intentionally stays with the substring rule.
2. packages/microcosm-build/src/microcosm/build/ledger_targets.py — one-token fix: _is_period_token now strips the ay prefix alongside ty/cy/fy. The SLC series key one record set per academic year (…ay2013 … …ay2029); without this the per-year record sets defeat the period-invariant key, the latest-at-or-before tiebreak refuses, and all ten SLC ids landed in multi_fact instead of activating. New unit test test__given_academic_year_record_sets__then_latest_source_period_is_used locks the behaviour.
3. Regenerated target_references.json (+11 references) and target_reference_membership.json: active 397 → 408, no_fact_at_or_before_period 18 → 7, multi_fact stays 1 (the documented genuine ONS residue). Reference-level diff verified: 11 added, 0 removed, 0 existing references changed — the ay normalisation perturbs nothing already active.
4. Count pins updated in test_uk_target_references.py, test_uk_national_calibration.py, test_country_spec.py; the Pin the scotgov council-tax stock targets to Scotland (activates the nine held CTAXBASE references) #754 scoped-pin test is rewritten as test_prefix_geography_pins_carry_scotgov_and_england_scoped_slc_families, which also asserts the seven deliberate holds stay held.

Verification

• Regenerated from the same pinned feed as Pin the scotgov council-tax stock targets to Scotland (activates the nine held CTAXBASE references) #754 (byte-identical reproduction recipe in that PR).
• New actives resolve as expected: SCP → fy2025 £471m projection; borrowers → AY2025 projections; support → AY2024 observations (e.g. maintenance-loan recipients 1,159,761) with uprating holds 2024→2025 registered alongside the existing slc.repayments.england_* ones.
• Full packages/microcosm-build pytest suite green locally; ruff format/check clean.

🤖 Generated with Claude Code