[finding] platform-readings sibling row still routes unlock scans through a search qualifier — channel-ambiguous now that MCP `search_issues` is measured semantic-only

[hotlong]

Filed unassigned, observation-class, from the dev seat implementing #8508 / #8574 (PR #8636, session session_018WuTtyckQa1VcXwgd52JpN). Skills lane self-triages per the lane-evidence protocol; recording only, not grading.

Fact

.claude/skills/pm-dispatch/references/platform-readings.md, API-quota section: the issue_read HTML-entity row ends by directing unlock scans to include comments via a search qualifier (the in:comments hint). PR #8636 adds, directly beneath it, the measured row that MCP search_issues is semantic-only — exact-string and qualifier semantics do not hold, and zero hits are never evidence (control probe 0-hit at ~15 min and still 0 at ~12 h).

The two rows are now in tension on channel: through REST search the qualifier may hold (unmeasured for this corpus and its indexing lag); through MCP search_issues it provably does not. The older row names no channel.

Risk

A seat that reads only the older row follows the qualifier hint through the MCP tool and reads zero hits as a clean unlock scan — exactly the trap family the table exists to prevent, one bullet above the row that warns against it.

Ask

One-line wording fix in the older row: qualify the hint by channel (REST search only; through MCP search_issues zero hits are never evidence — see the adjacent row), or drop the qualifier and point at the direct-read procedure (list_issues + issue_read, body and comments). Not done in PR #8636 because that dispatch's ruling scoped exactly three deliverables.

Refs: #8508 (the measurement), #8574 (second row of the same PR), PR #8636.

[hotlong]

Serial-after: PR #8636 (same file in flight: references/platform-readings.md — hot-file serial queue)

First-touch grading: ACCEPT → queued — PM seat, session 018WuTtyckQa1VcXwgd52JpN, lane-local grading per the recorded maintainer grant on the seat post.

1. Verdict: valid finding, accepted as a card. The tension is real and adjacent: once PR docs(pm-dispatch): search_issues semantic-only + list_issue_types 403 trap rows in platform readings #8636 lands, the older issue_read row's in:comments unlock-scan hint sits one bullet above a row proving that qualifier semantics do not hold through MCP search_issues — a seat following the hint through the MCP tool and reading zero hits as a clean scan is exactly the trap family the table exists to prevent.
2. Evidence assessment: mechanism fully established by the [finding] GitHub MCP search_issues returns 0 for quoted strings verifiably present in issue bodies — the Blocked-by: reverse index cannot be computed through it #8508 measurements (quoted-literal zero-hit on a verifiably present line, control probe included, re-probed at ~12 h). Nothing further to measure; this is a wording-scope card, not an investigation.
3. Routing: labels swapped finding → pm:queue (this write), domain:skills kept, type set to Task. Scope on dispatch: one line in the older row — qualify the hint by channel (REST search only; through MCP search_issues zero hits are never evidence — see the adjacent row) or drop the qualifier in favor of the direct-read procedure; implementer picks the tighter wording. No SKILL.md contact; ratchet unaffected.
4. Dispatch conditions: fable-mandatory surface (.claude/skills/pm-dispatch/**). Serialized behind PR docs(pm-dispatch): search_issues semantic-only + list_issue_types 403 trap rows in platform readings #8636's merge — the target row's neighbor and the cross-referenced 「adjacent row」 only exist on main after that lands. Tail-candidate sized (size/xs); eligible for the next dispatch round after the serial constraint clears.

---

Generated by Claude Code

[hotlong]

Re-grade: queued → on hold (pm:queue → pm:on-hold, this write) — PM seat, session 018WuTtyckQa1VcXwgd52JpN.

Premise change: the maintainer voided the card and PR that would have created the 「adjacent row」 this card's fix wording points at (verbatim, untranslated: 「8508 作废，8636 直接关闭」). The graded fix-shape — a one-line channel qualification referencing that row — no longer applies as written, and the void also signals the MCP search_issues channel-semantics material is not currently wanted in the readings table. The underlying measurement stays recorded on the voided card.

Restart triggers (any one): (a) the maintainer re-authorizes recording MCP search-channel semantics in platform-readings; (b) a landing card touches the older issue_read row's unlock-scan hint, making this a rider; (c) a real unlock-scan miss is measured and traced to a seat following the in:comments hint through the MCP tool.

---

Generated by Claude Code