UK: childcare take-up at the spring term, under-2 take-up and single-age targets behind the £100k childcare costing (policyengine-uk 2.147.2)

changelog.d/uk-childcare-dfe-single-ages.added.md

@@ -0,0 +1 @@
+The national UK contract binds four more DfE funded-childcare counts by single year of age, England, from the January 2025 census (María's ruling of 2026-10-21: bind every DfE childcare count the model can carry, even on few records, and sign out only those with a real reason; microcosm#1278, #1095): working parent registrations at ages 1, 2 and 3 (`dfe.funded_childcare.working_parent_children_age_<n>`) and universal registrations in nursery at age 3 (`universal_nursery_children_age_3`). Each counts persons in English households with whole-year age `n`; from age 2 the rows leave out children in primary education, as DfE's registrations leave out reception. The working parent 2 to 4 row (the take-up fitter's anchor) and the universal-only row stay bound beside them, unaffected. The contract's accounting notes sign out the rest of the table, each for its reason: the three age-4 cells, the working parent and universal nursery registrations at age 4 and the 4-year-olds in reception (DfE's January census reports ages as at 31 December, when about two-thirds of funded 4-year-olds are already in reception, while the model's whole-year age is the age at interview spread over the survey year, with fewer 4-year-olds in primary education, so binding the single age-4 cells would move weight between 4-year-olds in school and in nursery because of the reference date, not the data); the 9 to 11 months cell (policyengine-uk gives whole-year age 0 no working parent hours); reception at age 3 (no FRS 3-year-old is in primary education); the totals and multi-age cells that sum or copy bound cells or include an uncarried or signed-out cell; the universal all-provision cells (nursery plus reception); DfE's eligible-children and all-children estimates (built from population and survey-based inputs); and the registered percentages. The contract has 436 targets (432 before) and the registry scope 377 (373): four more contract-only declarations (191). A signed `in_year_snapshot` hold covers the four rows, the national references give 1,337 active references (1,333), and the compile-parity tool signs the four as `ledger_only`, with their rationale.

changelog.d/uk-childcare-dfe-spring-term.changed.md

@@ -0,0 +1 @@
+The DfE funded-childcare rows are measured at the spring term, as DfE's January census counts them: each row counts the children funded in the term that starts on 1 January of the model year's April-to-March span, by their age on 31 December, through four person measures (`uk_runtime/dfe_spring_term.py`) built on policyengine-uk 2.147.2's date-of-birth school start and working-parent term start. No row reads `current_education`. The universal and early-learning measures count every child of the DfE age, as the term after the birthday does, where the engine still gates them on the age on 6 October; DfE's rule of one entitlement per child holds at age 2. The working parent registrations at 9 to 11 months (January 2025) are bound with their own approved uprating hold, beside ages 1, 2 and 3, universal nursery age 3, and the working parent 2 to 4, early learning and universal-only rows, which also move to the spring term; the age-4 cells stay signed out, as the bound rows fix them. The ONS single-year note names the mid-2025 estimates' release date (1 October 2026).

changelog.d/uk-childcare-nested-rate-guard.changed.md

@@ -0,0 +1 @@
+The UK take-up contract refuses a nested rate (`nested_under`) unless its fitting receipt is from childcare fitter tool version 5 or later, the first to fit the conditional rate; an earlier receipt fitted the marginal rate.

changelog.d/uk-childcare-ons-ages-0-6.added.md

@@ -0,0 +1 @@
+The national UK contract binds eleven more ONS single years of age for all persons, for the £100k childcare costing fixes (item 6; microcosm#1278, #1095): the UK's ages 3 to 6, so the UK rows run from age 0 to age 6, and England's ages 0 to 6 (`ons.population.england_age_<n>`, pinned to E92000001, counting persons in households with `country == ENGLAND`). With ages 0 to 2 bound, calibration still moved the single years inside the 0-4 and 5-9 sex bands, holding too many 5-year-olds and too few 4-year-olds, and England's funded childcare is costed on England's children. Each row reads the ONS mid-2025 estimate by single year of age, the series the sex bands and the region 0-9 cells sum, and counts people with `age >= n` and `age < n + 1`; every band stays bound beside them. Wales, Scotland and Northern Ireland get no single-year rows, because each nation's single year rests on too few FRS households; the UK rows bind their sum, and Scotland's age 0 keeps its own row. The contract has 432 targets (421 before) and the registry scope 373 (362): eleven more contract-only declarations (187) beside the same 186 mapped targets. The national references and membership give 1,333 active references (1,322), and the compile-parity tool signs the eleven rows as `ledger_only`, with their rationale.

changelog.d/uk-childcare-take-up-nesting-and-copies.changed.md

@@ -0,0 +1 @@
+The UK childcare take-up draws now nest the universal 15 hours under the working-parent entitlement, re-draw the copies' childcare take-up, and fit and calibrate the DfE childcare rows without children in reception. `frs_take_up` keeps the extended draw unchanged and draws `would_claim_universal_childcare` as `would_claim_extended_childcare` OR u < the contract's `universal_childcare` rate, which is now the rate among benefit units not claiming the extended hours (the operation declares `nested_under`, and the stage refuses a manifest that drops it). Independent draws let a family over the £100k limit hold extended-yes and universal-no, and policyengine-uk then allocated its child's first 15 hours to the universal entitlement only once the limit was removed. The take-up signal gate targets the nested marginal, r_e + (1 - r_e) * r_u, and fails on any benefit unit claiming the extended hours without the universal ones. A new spine stage, `childcare_copy_take_up`, which runs after every copy stage and before `private_school_attendance` (whose net-income ranking counts the childcare entitlements), re-draws `would_claim_tfc`, the extended, universal (nested) and targeted flags and `maximum_extended_childcare_hours_usage` for the SPI support, high-value home support and CGT copies on their own `benunit_id`, through `frs_take_up`'s helper at the same rates, seed and salts, so a copy no longer inherits its host's draws; native units keep theirs, and the anchored flags and `would_claim_uc_childcare` are untouched. The roster moves from 46 to 47 stages. `tools/fit_uk_childcare_takeup.py` (tool version 5) fits the universal rate on the conditional basis, E = r_e * sum(with extended) + (1 - r_e) * r_u * sum(without), with no allocated term, and leaves children in primary education out of the working-parent and universal-only rows, as DfE's January registrations leave out reception; the two calibration bindings carry the same `current_education != PRIMARY` filter. The contract's numeric rates are unchanged until the fitter refits them.

changelog.d/uk-childcare-under-2-usage.added.md

@@ -0,0 +1 @@
+Children under 2 take their own working-parent take-up, one rate per DfE age (the age on 31 December of the model year, as DfE counts registrations; computed without the engine from the stored age and months since the last birthday, as policyengine-uk places the date of birth). `frs_take_up` draws, for each child of DfE age 0 or 1 in a benefit unit that claims the working-parent hours, whether it uses them, at its age's contract rate (`extended_childcare_age_0_usage`, `extended_childcare_age_1_usage`; keyed on `person_id`, seed 0, salt the output); a child that does not gets `max_free_entitlement_hours_used` 0, so policyengine-uk funds it no hours, and every other person keeps the engine's default of 30. DfE registers far fewer of the youngest eligible children than of those aged 2 to 4, and fewer at 9 to 11 months than at age 1 (mostly parental leave), so one family rate fitted at 2 to 4, or one rate for both ages, funded the wrong number of them. `childcare_copy_take_up` re-draws the copies' children against their own unit's re-drawn flag. The take-up signal gate measures the share over the children of DfE age under 2 in claiming units against their ages' weighted rate; the column is allowed on the export surface. The tool-version-7 childcare refit fits both rates against DfE's working parent rows at 9 to 11 months and age 1, each child counted at its DfE age.

changelog.d/uk-policyengine-uk-2-147.changed.md

@@ -0,0 +1 @@
+The UK lock moves from policyengine-uk 2.145.0 to 2.147.2, the release that carries pe-uk#2570 (early years hours stop at school start; working-parent hours start the term after the minimum age), with policyengine-core still at 3.33.3: the uprating pins' stamp, the concept coverage (311 to 314 inputs), the coverage manifest, the H2 fixture, the cited versions and `APPROVED_UV_LOCK_SHA256` re-derive. The concept mapping's liquid-asset reason follows 2.147.0, where Universal Credit and Pension Credit count `share_like_capital`, which is `corporate_wealth` unless its components add up to more. The build leaves the three new inputs at their defaults: `directly_held_shares` and `unit_and_investment_trusts` (the WAS stage's `corporate_wealth` already holds the shares and funds, so `share_like_capital` equals it) and `early_years_school_start_deferred` (the FRS records no deferred or council-funded independent reception start, so every child starts school on the date its date of birth implies).

changelog.d/uk-war-widows-pension-export-surface.fixed.md

@@ -0,0 +1 @@
+The release export surface allows `war_widows_pension_reported`, which the FRS spine writes since #1222 and the enhanced FRS lacks (it still adds FRS benefit code 9 into `bsp_reported`); without it, release-cut certification blocked every build from main. A new engine-free test runs the export-surface gate on the spine's declared columns, so a stage output outside the enhanced-FRS inputs and the allow-list now fails CI instead of certification.