docs/MANIFEST.toml still cites R10.2/D17.9 — the citation sweep in #47 could not see it

[forkwright]

Finding

docs/MANIFEST.toml's entry for docs/design/rendering-completeness.md still describes the doc as the "R10.2/D17.9 rendering-completeness floor". #47 (commit 74f6622) established that R10.2 is the operator-facing-GUI requirement, has never described the rendering floor, and rewrote every other citing file (rendering-completeness.md itself, docs/design/topology.md, README.md) to cite D17.9 alone. docs/MANIFEST.toml carries the same stale pairing #47 fixed everywhere else, because it was never in scope for that sweep to find.

Evidence

docs/MANIFEST.toml:44:

description = "Measurement contract for the R10.2/D17.9 rendering-completeness floor over the origin corpus."

Compare docs/design/rendering-completeness.md:3 (fixed by #47): "This note defines the measurement contract for D17.9." and README.md:28 (fixed by #47): "D17.9 evidence floor".

Root cause: ci/check-doc-refs.py, the CI stage (doc-refs-check) that exists specifically to catch this identifier-citation class, only scans root.rglob("*.md"). docs/MANIFEST.toml is a .toml file, so its own description prose — despite carrying the exact same R<n>/D17.<n> citation shape the script's JOINED_IDS_RE is built to check — is invisible to it. ci/check-doc-manifest.py (docs-manifest-check) validates that manifest paths exist/are complete and that each file passes kanon lint --writing; it never inspects a manifest entry's own description text for citation accuracy either. So no stage in .kanon-ci.toml can currently catch this class of drift in docs/MANIFEST.toml itself.

Why this matters

docs/MANIFEST.toml is documented as the authoritative doc inventory ("makes docs/MANIFEST.toml the authoritative CI input set" per check-doc-manifest.py's own docstring) and is the first thing a reader or an agent consults to learn what each canonical doc is for. A stale citation there re-teaches the exact wrong pairing #47 spent an issue correcting everywhere else, and nothing in CI will catch a repeat of this specific failure mode (a citation fix that updates every .md file but not the manifest's own prose).

Desired correction

• Immediate: fix the one stale description (drop R10.2/, matching the D17.9-only wording used everywhere else #47 touched).
• Systemic: extend ci/check-doc-refs.py (or check-doc-manifest.py) so docs/MANIFEST.toml's description fields are checked by the same dangling/mismatch logic as every other canonical doc. Note this is not a trivial glob-widening: docs/MANIFEST.toml:33 describes docs/design/decisions.md as covering "D17.x technical decisions" — the bare-ref scanner would need to treat D17.x the way RANGE_CONTEXT_RE already treats D17.*, or a naive glob change introduces a false-positive dangling report on that entry. Add a case to check-doc-manifest.py --self-test (or an equivalent fixture) proving a re-injected stale citation in the manifest is caught, mirroring the existing self-test pattern.

Done when: docs/MANIFEST.toml:44 cites D17.9 only, and a CI stage exists that would fail if that citation regressed.