Input-data basis as a surfaced claim attribute (people assume EUROMOD means SILC)

Max (2026-08-29): "we need to be careful with euromod - people assume thats silc." The assumption is correct for the Belgian country-report claims — EUROMOD J1.0+ totals simulated on uprated EU-SILC 2022 (income year 2021, private households) — but the page only said so deep in row conditions until #117 added it to the BE prose.

Prose is the reform-specific fix. The scorecard-shaped fix is a first-class, machine-readable input-data basis on every external claim, rendered as a chip next to the source label, so no reader has to assume:

• model_on_survey — EUROMOD-BE on EU-SILC (already pinned per-row as the input_database condition be_2022_c1_silc2022_income2021); UKMOD similarly.
• model_on_admin_microdata — SPF Finances PIT-reform statement (DOC 56 1243/001 names assessment-year-2023 microdata; already in the publication notes).
• admin_outturn / forecast — already a per-row routing tag from the Chronicle boundary work.
• PE/Axiom side gets the same treatment (Microcosm basis per country — demo-grade BE reweighting is exactly the kind of thing the chip should say).

Most of the raw material already exists in conditions/publication JSON — this is promotion to a closed-vocabulary column plus display, not new research. The vocabulary needs one careful pass (survey vs admin vs mixed vs uprated-survey), and the display must stay descriptive: the chip states the basis, never grades it.

Related: #10 (assumptions registry — the basis chip is arguably its first shipped column), #13 (baselines precedent for promoting a buried attribute to first-class).

Start by tracing the existing conditions and publication JSON, including input_database and routing tags, to see how external claims reach the source-label display. Done means every external claim has a closed-vocabulary input-data basis rendered as a descriptive chip, with survey, administrative, outturn, and forecast distinctions handled consistently.