UK #882 repairs: the A16 legacy rows, the carer and childcare elements, the benefit-cap bands

changelog.d/882-uk-benefit-cap-bands.added.md

@@ -0,0 +1 @@
+Bind the fourteen DWP benefit-cap amount bands (Table 4, November 2025) at benefit-unit grain, the up-to-£100 band in the objective and the thirteen others measured on the reviewed exclusion register while the capped tail rests on a few cloned households, with `tools/diagnose_uk_benefit_cap.py` as the tail diagnostic (#882).

changelog.d/882-uk-carer-element.added.md

@@ -0,0 +1 @@
+Fill the engine's `care_hours` from the FRS adult HOURTOT band (lower band edge; codes 5, 6, 7 and 10 reach the 35-hour line) and set `would_claim_carers_allowance` from reported receipt in the root spine stage, on policyengine-uk 2.98.0 (floor raised from 2.97.0) whose `is_carer_for_benefits` recognises the hours, so the Universal Credit carer element reaches carers without a Carer's Allowance award while the allowance stays on reported claimants (#882). The carer element register row is retired on the twin evidence (paid-UC carer units +80 percent at design weights). The hours limb counts reported hours alone, with no check that the person cared for receives a qualifying benefit, as the engine documents.

changelog.d/882-uk-childcare-takeup.added.md

@@ -0,0 +1 @@
+Draw `would_claim_uc_childcare` in the `frs_take_up` stage at one contract rate per family type (single 0.441, couple 0.2682, fitted by `tools/fit_uk_uc_childcare_takeup.py` so the expected paid-UC count of the Universal Credit childcare element equals DWP's calendar-2025 single and couple households), carried on the extra export surface as a net-new engine input that policyengine-uk 2.98.0 reads (#882). The childcare element register row is retired on the twin evidence.

changelog.d/882-uk-legacy-rows.changed.md

@@ -0,0 +1 @@
+Re-adjudicate the A16 measure exclusions `obr.housing_benefit` and `dwp.jsa_claimants` on the mechanism receipts of `tools/diagnose_uk_legacy_benefits.py` (the SPI support channel carries reported Housing Benefit and JSA into synthetic households; the Universal Credit take-up draw zeroed Housing Benefit for pension-age units), gate the `would_claim_uc` draw to benefit units with an adult under State Pension age (the U8 rate is unchanged, its population re-signed), bind the DWP Housing Benefit caseload by tenure as measured diagnostics, and declare the ESA single-type and credits-only reconciliation on the three ESA rows (#882, #867, #869).

changelog.d/882-uk-register-names.fixed.md

@@ -0,0 +1 @@
+Key the three UC element rows on the UK measure-exclusion register by their reference names (`dwp.uc.households_{carer,childcare,housing}_element`) instead of their binding metric names, which matched no compiled spec and made the calibration run refuse; a test now requires every register name to resolve to a committed reference (#882).

docs/uk-uc-baseline-2026-09-10.md

@@ -90,3 +90,12 @@ Compile-time target values are calendar-2025 means of the Payment Indicator = Ye
 Deductions and the two tenure rows are within 6 % on both files and LCWRA within 15 % (Q50f) / 20 % (D2); those four rows are active. LCWRA stays in the objective by the rule its binding note states: a concept mismatch goes to the register only when it puts the row outside the 25 % release bound at the baseline. Carer and childcare are measured but held out of the objective on the reviewed exclusion register until their model gaps are repaired (see the element-support section above). The any-tenure housing row is also measured but held out: DWP's 'Yes' is social plus private plus 'Other or unknown' tenure, the model has no other/unknown tenure, so its any-tenure count is identically the two tenure rows and the published total exceeds what the model can reach by the other/unknown cell (112,518 in calendar 2025, 2.8 %); fitting all three would over-fill the tenure rows. The register entries are signed by María in review of the element PR (2026-09-15) with a one-month window. The paid-claim targets sit below the April–December all-claims means quoted in the earlier section because the calendar window includes the lower January–March 2025 months and excludes nil-payment claims.
 
 Paid GB caseload for scale: Q50f 6,416,983, D2 6,428,401; share of paid units with earnings 51.6 % / 49.6 %.
+
+## Repairs measured against this baseline (16 September 2026, the #882 repairs PR)
+
+Receipts: `experiments/882-uc-repairs-receipts.md`. Twins built with the licensed recipe at design weights: a control at the branch base (51c31438, engine 2.97.0) and the repaired branch (engine 2.98.0, PolicyEngine/policyengine-uk#1857). Projections onto Q50f use the design-weight ratio repaired / control.
+
+- Carer element: the engine's carer condition now holds on Carer's Allowance receipt or 35 weekly care hours, and the spine fills `care_hours` from the FRS adult HOURTOT band with `would_claim_carers_allowance` following reported receipt. Paid-UC carer-element units 425,757 → 768,273 (+80 %); projection 1,075,100 against 1,078,514 (−0.3 %). Carer's Allowance spend stays on reported claimants once the take-up flag is re-derived after the SPI fill. The register row is retired.
+- Childcare element: `would_claim_uc_childcare` is drawn per family type (single 0.441, couple 0.2682) so the expected paid-UC count equals the published calendar-2025 means at the Q50f weights; on the twin the realized count is 116,324 against 353,470 at rate one, below the expectation because SPI-synthetic units copy their donor's flag on a base of 378 and 201 records; projection 149,950 against 178,667 (−16 %). The register row is retired with that caveat recorded.
+- Legacy rows: the A16 exclusions `obr.housing_benefit` and `dwp.jsa_claimants` are re-adjudicated to 2026-12-08 on the mechanism receipts (the SPI support channel carries reported Housing Benefit and JSA into synthetic households; the Universal Credit draw had zeroed Housing Benefit for pension-age units). Gating the draw to units with an adult under State Pension age lifts engine Housing Benefit from 401,489 to 644,183 units at design weights (£2.61bn → £4.11bn) and moves paid UC −1.8 % (the rate now realizes 0.559 among the eligible units where the old all-units draw realized 0.582); three Housing Benefit caseload rows bind as measured diagnostics.
+- Benefit-cap amount bands: fourteen DWP Table 4 bands bind at benefit-unit grain; the up-to-£100 band is in the objective and the thirteen others are measured on the register. On Q50f the tail above £1,300 a month is 14.2 % against 0.6 % published; on the spine at design weights it is 6.6 % from four source households (eight records), so the Q50f tail is a selection-and-calibration effect on a thin base.