Column closure: every artifact column claimed by exactly one lineage class (spec totality)

Max, 2026-08-15: the process diagrams should have revealed the geography flaw sooner — the observability has to be total, not just cover the parts we extracted.

The measurement

Column-closure audit of the f025 artifact against the lineage spec (#695), 2026-08-15:

• 392 columns across 6 entities; spec claims 156 (+10 structural) ≈ 42%; 226 unclaimed.
• In the unclaimed dark matter: ~187 raw source (caps) columns; the 3 geography columns (where the #696 flaw hid); ~12 takes_up_* flags — stochastic draws with policy consequences, declared nowhere; capital-gains-tail donor receipts; spine-source ids and role/filing inputs.

Inventory: one JSON fixture of entity/column names extracted from the artifact (names only, no data).

The increment

1. Total lineage classes in specs/us_imputation_lineage.yaml: measured_native (per source channel), imputed (existing), computed (existing), assigned_by_ladder (#696 lane adds), take_up_draw (new — each flag's distribution, conditioning, parameter source, and calibration status), receipt, structural.
2. Column-closure conformance test: the artifact column inventory is a committed fixture; every column must be claimed by exactly one class entry; unclaimed or doubly-claimed ⇒ test failure. Inventory regenerated by a tool from any built artifact; drift between fixture and a fresh build fails loudly.
3. Coverage meter on /lineage: claimed/total rendered in the hero, per class — the number that keeps us honest.
4. Immediate deep-dive owed: the take_up_* mechanism (where drawn, from what) documented into the spec as part of this increment — it is currently exactly as invisible as geography was.

Sequencing: lands after the two wave-1 lanes (#696 block-first, per-family predictor sets) merge, since all three edit the same spec file. Owner runs it as its own PR.

Start with specs/us_imputation_lineage.yaml and the committed artifact column inventory, then locate the tool that regenerates the inventory from a built artifact and the /lineage rendering. Run the column-closure conformance test first. Done means every artifact column has exactly one lineage class, fixture/build drift fails, take_up_* behavior is documented, and coverage is shown per class.