DR-153: Gap visibility is a platform surface, not bundle content

Context

DR-141’s coverage guarantee was implemented (id-427) with a per-document residual grain: any published source document with no published q_a_pair derived from it receives a template concept (“no published answer has been distilled from this document”). The S564 bundle inspection measured the outcome on the staged S561 bundle: 32 of 49 concepts (65%) are content-free boilerplate, ~19 of them CI test fixtures admitted to the corpus, with one publishing fixture text as knowledge; documents/ is larger than every other subdirectory combined, and the boilerplate dominates both indexes. The owner’s scaling test is decisive: a client with 10,000 documents would receive 10,000 stub files drowning the actual knowledge. The grain also re-imports the pre-reframe model into the bundle — it treats evidence/register entries (corpus reframe A/B) as units of knowledge (C).

Decision

Owner-ruled (S564 board, D5-A): per-document residual concepts retire from the bundle. A source document with no published knowledge derived from it gets no concept file. Corpus gap visibility — the genuinely valuable intent behind the grain — moves to a platform coverage surface over the provenance register (what was ingested; what knowledge, if any, was admitted from it), which is also the corpus reframe’s named week-1 gap-surfacing story. The bundle may carry at most a single gaps index, never one file per document. DR-141’s coverage guarantee stands, honoured at the knowledge grain: every published unit of knowledge remains reachable in at least one concept.

Alternatives Considered

• Keep the grain, fix hygiene (unpublish test fixtures, real corpus only) — rejected as the ruling: hygiene removes this bundle’s noise but not the scaling inversion, and the grain’s premise (register entries are concept-worthy) is the reframe violation.
• Keep the grain but suppress it from indexes — display-side concealment of content that should not exist; rejected unconsidered-in-detail.

Consequences

• The rebased producer (id-465 / DR-152) implements enumeration without the document residual; the retirement executes there, not as a patch to the current tree.
• id-427’s residual-cascade implementation and its tests do not port.
• id-462’s fixture-unpublish hygiene remains correct and independent (test scaffolding must not be published corpus regardless of grain).
• The platform coverage surface (register-backed gaps view) needs an owning task when consumption-surface work begins; until then the requirement is recorded here.