[finding] platform-objects' `packages/**/*.object.ts` is held on the roster by an exclusion-list literal, not by the walk that really holds it

[claude]

Observation-class, measured while building the per-glob holder limb in PR #10690. No gate is red today, and this is not a defect — it is a green that is right for the wrong reason. Filed unassigned, no pm:queue.

Measured on 844bf29a84

The new per-glob limb asks whether each declared glob still holds something on its package's roster. @objectstack/platform-objects passes:

### @objectstack/platform-objects (escaping: 1 test(s))
 roster literals: 1
 HELD packages/**/*.object.ts <- packages/create-objectstack/src/templates/blank/src/objects/note.object.ts

That single roster path is a quoted literal in an exclusion list inside packages/platform-objects/src/managed-api-method-affordance-sweep.test.ts (:141), naming a template object the sweep deliberately skips. What actually holds the glob is the test's walkObjectFiles(PACKAGES_DIR) walk — which descends with readdirSync(dir) on a loop variable, so the detector resolves the escape and loses every name.

The two other packages declaring the same glob (@objectstack/spec, @objectstack/plugin-auth) have no such incidental literal, which is why they needed the heldBy witnesses PR #10690 adds and this one did not.

Why record it

The mechanical holder and the real holder are different things here. Deleting or rewording that exclusion entry — an ordinary edit, nothing to do with the radius — flips this glob to roster-invisible. The consequence is benign by construction (the gate then asks for a heldBy witness by name, which is the correct prompt), so nothing is broken; what is worth recording is that the green is currently supplied by a line whose purpose is unrelated, the same shape as #9763's "prose was holding the radius" one register down.

The cheapest treatment, if anyone wants one: declare the walker as this glob's heldBy witness alongside the other two packages, so the mechanical holder and the stated holder agree. Not done in PR #10690 on purpose — that PR declares witnesses only where the tree needs them to stay green, and widening the change to globs that already pass would have made the six load-bearing declarations harder to review.

Refs

#10566 (the limb) · PR #10690 · #10315 (derived paths a source scan cannot see) · #9763

---

Generated by Claude Code

[os-zhuang]

Triage (finding concentration round, seat grading): closing not planned. The failure mode is self-correcting by the gate's own design — if the incidental exclusion-list literal ever stops holding the glob, the per-glob limb prompts for a heldBy witness by name, which is the correct outcome, not a silent gap. The card is the record; the cheapest treatment (declare the walker as heldBy alongside the other two packages) is a one-line rider any future PR touching managed-api-method-affordance-sweep.test.ts or the limb may take. Maintainer may veto/reopen; listed in the round report.

---

Generated by Claude Code