UK UC deductions: external scores staged (FRR, JRF floor, historical cap costings)

PolicyEngine UK grew a UC deductions module today (policyengine-uk#1815, linked to policyengine-uk#1814), which opens a new UK comparison family. External claims are harvested, verified against primary sources, and staged at ~/scorecard-harvest/uk_deductions/ (CLAIMS.md = collation; source PDFs/text alongside; every figure page-verified 2026-08-05).

Claim families staged

1. Fair Repayment Rate (cap 25%→15%, 30 Apr 2025) — HMT AB2024 (1.2m households, £420/yr avg; PSNCR +£385m 2029-30), DWP official statistics Nov 2024 (1.2m families with country splits, 3.0m individuals, May 2024 caseload, OBR-certified methodology), DWP/DfC screening (5,400 losing households, 22k child-maintenance behavioral), Policy in Practice per-household-type maximums.
   • PE comparator already computed (PR #1815, enhanced FRS 2023/24): 0.85m benunits better off, mean annual gain £421 (DWP: £420), £0.36bn/yr deductions reduced.
2. JRF protected minimum floor — Oct 2024 pre-FRR figures (1.9m families/£48/mo/£150m, JRF's own modelling, no version named) vs Oct 2025 post-FRR figures (IPPR TBM v02_78: 130k out of poverty, 90k families £75/wk, £340m→£380m; £800m with two-child-limit removal).
3. Historical cap changes — Budget 2018 (40%→30%) and Budget 2020 (30%→25% + 24-month advances) financial-transactions costing lines, year by year. These score directly against PE's dated cap parameter history.
4. Other — Citizens Advice advances-to-grants costings (£0.75–1.49bn/yr), CPAG/StepChange scale estimates.

Hygiene notes for ingestion

• Route outturns to Ledger, not external_scores: the DWP deductions statistics (quarterly tables, Stat-Xplore) are the calibration source PE's parameters consume — calibration_relationship: consumed_as_target. PE-vs-those-tables rows are the tautology the boundary rule exists for. The module's parameter provenance cites the March 2025–February 2026 supplementary tables specifically.
• FRR fiscal claims: benchmark against PSNCR, never PSNB. Verified absences: no FRR line in HMT Table 5.1, the AB2024 policy costings document, or the Oct 2024 OBR EFO. PQ UIN 3751 (Jun 2026) states no PSNB impact and points to the £385m PSNCR figure. A PSNB comparison row would compare against a number that does not exist.
• JRF floor needs baseline_key discipline: pre-FRR (2024/25) and post-FRR (2025/26) variants differ by an order of magnitude in cost for the same nominal policy. They are different baselines, not a revision.
• Concept mismatch flag for floor rows: JRF's floor binds on deductions and benefit-cap reductions; post-FRR its bite is mostly benefit-cap interaction. PE's deductions cap alone is not the same reform — a comparable run applies the floor jointly. Until that run exists, floor rows are constructed/concept_mismatch, not comparable.
• Attribution correction recorded: an earlier session note attributed JRF's Oct 2024 floor to IPPR TBM v01_57 — unverifiable; v02_78 is named only on the Oct 2025 briefing. policyengine-uk#1814 carries the correction.

Not ingesting to the DB from this session (one-writer coordination); this issue + the staged collation are the handoff for the next ingest lane.

🤖 Generated with Claude Code

Start with ~/scorecard-harvest/uk_deductions/CLAIMS.md and the staged source files, then inspect the scorecard's existing ingestion and external-score entry points. Confirm the FRR PSNCR benchmark, baseline_key distinctions, and floor concept-mismatch rules before deciding what can be ingested; done means the staged claims are routed and classified without tautological calibration rows.