Reform-period-metric view: every source's score for one target in one place

Max (2026-08-29, from the Belgium PIT-reform lane): "we should be able to see all scores for a given reform-period-metric in one place."

Reform validation today is row-per-claim, keyed by source. The Belgium case makes the gap concrete: three sources publish a revenue change for the same reform at the 2030 horizon — SPF Finances −€4.0B, Cour des comptes −€5.0B, PolicyEngine −€4.16B — and they render as three separate table rows, with the period-basis caveats repeated per row instead of stated once for the group.

Ask: a grouped surface keyed by (reform, period, metric) that renders every source's value side by side.

Existing analogues this would serve immediately:

• BE: the three PIT-reform 2030 estimates above (period bases differ and stay labeled unresolved — the group view must carry that per pair, not resolve it).
• US: JCT / CBO / TPC / PWBM scoring the same provision and window (the OBBBA triangle; the two-child-limit pentagon is the UK cousin).
• UK: HMRC ready-reckoner vs PE per tax head and forecast year.

Design notes from the registry as it stands:

• The grouping key already exists in the DB: ReformRef + period + metric (+ unit concept). No new ingest needed for a first cut.
• Cross-source display needs pairwise status/benchmark-class semantics — today's benchmark_class/status is PE-vs-external; source-vs-source cells in a group view are constructions too and need the same descriptive labeling (no "closest", no implied winner — doctrine #9).
• Baselines differ across sources (#13 registry) — a group view must show each member's baseline, or mismatched-baseline members render as plain disagreement.
• Related but distinct: #10 (assumptions registry: why they differ), #9 (app views). This issue is the where they all are surface.

Start by tracing the existing row-per-claim rendering and the registry grouping key of ReformRef, period, metric, and unit concept. Review benchmark_class/status and baseline fields, then define a grouped surface that places source values side by side while preserving pairwise labels, per-source baselines, and unresolved period-basis caveats without implying a winner.