HMT distributional analysis (mode 2, distributional): decile-impact counterparts to the Budget 'impact on households' tables

Every fiscal event ships HM Treasury's Impact on households: distributional analysis — decile-level cash and %-of-income effects of the announced package, plus the tax/welfare/public-services split. The #32 harvest staged the HMT lane (1,368 claims, per the #3 fleet dispatch: Budget costings docs + distributional analysis). Nothing turns it into comparisons: no issue, no registry, no PE counterpart. Yet this is the population that exercises what PolicyEngine is for — household-level incidence — and the UK analogue of the US side's TPC/Urban distributional tables, which have been populated since day one (#11, #12, #31).

What makes it a distinct lane (not #54/#60)

Aggregate costings validate revenue totals; this validates the shape — who gains and loses by decile. PE-UK computes decile impacts natively, so the counterpart compute is cheap; the work is in the axes:

• Package construction: HMT scores the cumulative announced package against a counterfactual, not single measures — the registry entry is a package (a composition of #54 registry measures), which is exactly what the measure→reform registry's construction field was designed to hold.
• Income concept + equivalisation: HMT uses its own equivalised net-income deciles and includes public-service spending in some presentations; the tax/welfare-only split is the comparable slice. Axes go to the assumptions registry (#10).
• Basis: HMT DA is static-ish (closer to PE than OBR costings) — a better-matched benchmark class, worth recording per row.

Steps

• Adapter for the staged HMT DA tables (AS2023 → AB2025): decile × measure-group × {cash, % net income} claims with full conditions (fiscal event, counterfactual baseline as published, income concept)
• Package registry entries composing #54 measures per fiscal event, computability triage for the non-tax-benefit components (public services → not_expressible, labelled)
• PE-UK decile counterparts on the certified bundle via the #56 pipeline pattern (household decile ranking recorded as an artifact so the ranking concept is auditable)
• Descriptive comparison: per-decile signed gap, named axes, no summary score; benchmark_class = different_model

Depends on #57 (ingest shape) and #54's registry (package composition). Complements #59 — a package whose total matches but whose deciles don't is a different defect family than a costing miss.