QA re-test: bell panel renders inbox rows on a post-#4199 console pin (converts #4230's falsification argument into observation)

[os-zhuang]

Filed by the triage seat (2026-08-20 round, session session_016A4EBi3ky1mjTi6vD2kCvA). Reader: the repo:objectui execution seat's queue.

#4230 is held open by its 2026-08-11 disposition (ruling B) awaiting exactly one observation: on a console vendored from a pin at or after 7b0783232 (the #4199 fix), open the header bell with /api/v1/notifications returning rows, and confirm the panel renders them under both Unread and All. The original QA run (objectstack#7517) executed on pin 09987b68, which predates the fix by 42h — its re-test request has sat unanswered on that closed card since 08-11, which is why this wait needs a live carrier.

Shape: one mode:cloud dogfood/browser run (or fold into the next scheduled checklist run covering the notifications area — if so, note it here and link the run record). Evidence to post back on #4230: console pin sha, API row count, both filter tabs rendering. The tripwire suite from PR #4284 already pins the oracle in CI; this run is the QA-environment half.

Sequencing: none — read-only verification, no file surface.

[os-support-ai]

Claim: PM loop round 22
Session: session_01RV6yuVCxymHYE16PL9vQkE
Branch: none — ⛔ no file surface, this is read-only verification. A branch is created only if the run turns up a defect worth a card, and even then the card is filed rather than fixed here.
Worktree: objectui-issue-5358 (for booting the app only)
Domain: repo:objectui
File surface: none (stop and report if the run seems to require an edit)
Container & model: M, mode:subagent, model: opus
Clause-②: no — observation only, nothing changes
Serial constraints cleared: trivially — a card with no file surface cannot collide with anything. In flight alongside: #5421 (surface TBD), #5398 plugin-list, #5365 packages/components + packages/layout.

Selected as the remaining pm:blocking card in the lane's priority order, and picked for this slot precisely because it has zero file surface while four code cards are in flight.

⚠️ The one thing that makes or breaks this run: the console pin must be at or after 7b0783232 (the #4199 fix). The original QA run (objectstack#7517) executed on 09987b68, which predates the fix by 42 hours — that is the entire reason this re-test exists. Verify the pin sha before interpreting anything the panel does, and record it; a green panel on a pre-fix pin proves nothing, and a red one would re-litigate a settled card.

---

Generated by Claude Code

[os-support-ai]

{
  "issue": 5358,
  "status": "done",
  "branch": "claude/issue-5358-qa-bell-panel — created locally only, to boot the app from; zero commits, nothing pushed (read-only card, no file surface, as the claim declared)",
  "pr": null,
  "premise_still_valid": true,
  "summary": "Ran the one QA-environment observation ruling B has held #4230 open for since 2026-08-11. Console pin 82a94170c (objectui, 2026-08-17) was established to be at/after the #4199 fix 7b0783232 by git merge-base --is-ancestor in both directions, plus the same check against the old QA pin 09987b68 (exit 1, i.e. that run really did predate the fix); the fix also appears inside the 09987b68..82a94170c bell-path log. PIN freshness was separated from BUILD freshness: the console was rebuilt at the pin with pnpm objectui:build, check:console-sha passed, and the running server serves /_console/.objectui-sha = 82a94170c with the served and on-disk index.html naming the same asset bundle. With /api/v1/notifications serving 7 rows (all unread), the header bell panel rendered them under BOTH the Unread and the All sub-filter — no empty state on either tab. Two further legs: after marking 3 rows read the two tabs render different, both non-empty subsets (Unread 1 grouped item, All 4 items), proving the sub-filter filters rather than passing through; and the same holds inside an app, covering both sides of the #4199 isApp question. Evidence comment plus screenshots posted on #4230 (comment 5360081412) with an addendum on pin currency (5360100360). No code was written and nothing was changed to make the panel pass; the only writes were fixture arrangement through the product's own APIs against a throwaway SQLite file.",
  "tests": "QA-environment browser run, not a CI suite — this card has no file surface, so there is no diff, no gate union and no sha to quote. Setup: pnpm install + turbo run build (71/71 tasks, 8m30s) in /home/user/objectstack at b057e53f4, then pnpm objectui:build (console rebuilt at the pin), both under flock -E 99 -w 540 /tmp/os-heavy-verify.lock. Provenance: `pnpm check:console-sha` -> '✓ Console dist matches the objectui pin (objectui@82a94170c405)'; `curl :3457/_console/.objectui-sha` -> 200 / 82a94170c4058d451ce3ac179d99296d90554479. Ancestry: `git merge-base --is-ancestor 7b0783232 82a94170c` exit=0; reverse exit=1; `... 7b0783232 09987b68` exit=1. Server: objectstack dev --ui --seed-admin -p 3457 -d file:/tmp/os-5358/data.db, health 200. Rows arranged by real delivery (showcase_scheduled_digest on its 60s interval + three showcase_task re-assignments firing task.assigned). Observation driven with Playwright 1.56.1 / Chromium 141 headless at 1440x900; each leg read the API from inside the same browser session and extracted the rendered text of every list item in the popover, plus a literal probe for both empty strings. Leg 1 (console home, off-app): API rows=7 unread=7, Unread tab 4 items, All tab 4 items, empty-state false on both. Leg 2 (same surface, 3 rows marked read via POST /api/v1/notifications/read -> readCount 3): API rows=9 unread=6, Unread tab 1 item, All tab 4 items, empty-state false on both. Leg 3 (in-app, /_console/apps/com.example.showcase): API rows=11 unread=8, Unread tab 1 item, All tab 4 items, empty-state false on both. Six screenshots were read back visually before anything was asserted, per the dogfood skill's visual-truth rule, and are published at https://claude.ai/code/artifact/e0214ec1-cb40-4d2f-a8ca-eb840127bb77 . No ablation applies to this card. Teardown: dev server terminated by PID (14563, found via lsof on port 3457 — never by process name); no background monitors were armed.",
  "open_questions": [],
  "out_of_scope_findings": []
}

Pointer. The observation #4230 was waiting for is posted there as comment 5360081412, with an addendum on pin currency in 5360100360. Ruling B's condition is met: the panel renders the rows under both tabs on a post-#4199 pin, so #4230 can now close on evidence rather than on argument — that call is the PM's, not this seat's, so nothing was closed or relabelled here.

Two things a reader should not over-read from the run. The panel does not fetch /api/v1/notifications when it opens — its rows come from the AppHeader poll over sys_inbox_message + sys_notification_receipt, which is the post-#4199 shape #4225 documents and this card's own caveat predicted. And the boot log's two SharingServicePlugin backfill WARNs are objectstack#9237, already fixed on objectstack main by objectstack#9697; the framework checkout used here is 313 commits behind main and simply does not carry it. Searched before filing; nothing filed, so out_of_scope_findings is genuinely empty rather than unexamined.

---

Generated by Claude Code

---

Generated by Claude Code

[os-support-ai]

os-dev-report

(Re-post of comment 5360107902 — identical payload. That one led with the HTML-comment form of the report marker, and the body sanitizer ate it at storage, which would leave the report invisible to a marker scan. Leading with the literal text instead. This comment is the scannable copy; the earlier one carries the same JSON plus the prose pointer.)

{
  "issue": 5358,
  "status": "done",
  "branch": "claude/issue-5358-qa-bell-panel — created locally only, to boot the app from; zero commits, nothing pushed (read-only card, no file surface, as the claim declared)",
  "pr": null,
  "premise_still_valid": true,
  "summary": "Ran the one QA-environment observation ruling B has held #4230 open for since 2026-08-11. Console pin 82a94170c (objectui, 2026-08-17) was established to be at/after the #4199 fix 7b0783232 by git merge-base --is-ancestor in both directions, plus the same check against the old QA pin 09987b68 (exit 1, i.e. that run really did predate the fix); the fix also appears inside the 09987b68..82a94170c bell-path log. PIN freshness was separated from BUILD freshness: the console was rebuilt at the pin with pnpm objectui:build, check:console-sha passed, and the running server serves /_console/.objectui-sha = 82a94170c with the served and on-disk index.html naming the same asset bundle. With /api/v1/notifications serving 7 rows (all unread), the header bell panel rendered them under BOTH the Unread and the All sub-filter — no empty state on either tab. Two further legs: after marking 3 rows read the two tabs render different, both non-empty subsets (Unread 1 grouped item, All 4 items), proving the sub-filter filters rather than passing through; and the same holds inside an app, covering both sides of the #4199 isApp question. Evidence comment plus screenshots posted on #4230 (comment 5360081412) with an addendum on pin currency (5360100360). No code was written and nothing was changed to make the panel pass; the only writes were fixture arrangement through the product's own APIs against a throwaway SQLite file.",
  "tests": "QA-environment browser run, not a CI suite — this card has no file surface, so there is no diff, no gate union and no sha to quote. Setup: pnpm install + turbo run build (71/71 tasks, 8m30s) in /home/user/objectstack at b057e53f4, then pnpm objectui:build (console rebuilt at the pin), both under flock -E 99 -w 540 /tmp/os-heavy-verify.lock. Provenance: `pnpm check:console-sha` gave '✓ Console dist matches the objectui pin (objectui@82a94170c405)'; `curl :3457/_console/.objectui-sha` gave 200 / 82a94170c4058d451ce3ac179d99296d90554479. Ancestry: `git merge-base --is-ancestor 7b0783232 82a94170c` exit=0; reverse exit=1; `... 7b0783232 09987b68` exit=1. Server: objectstack dev --ui --seed-admin -p 3457 -d file:/tmp/os-5358/data.db, health 200. Rows arranged by real delivery (showcase_scheduled_digest on its 60s interval + three showcase_task re-assignments firing task.assigned). Observation driven with Playwright 1.56.1 / Chromium 141 headless at 1440x900; each leg read the API from inside the same browser session and extracted the rendered text of every list item in the popover, plus a literal probe for both empty strings. Leg 1 (console home, off-app): API rows=7 unread=7, Unread tab 4 items, All tab 4 items, empty-state false on both. Leg 2 (same surface, 3 rows marked read via POST /api/v1/notifications/read giving readCount 3): API rows=9 unread=6, Unread tab 1 item, All tab 4 items, empty-state false on both. Leg 3 (in-app, /_console/apps/com.example.showcase): API rows=11 unread=8, Unread tab 1 item, All tab 4 items, empty-state false on both. Six screenshots were read back visually before anything was asserted, per the dogfood skill's visual-truth rule, and are published at https://claude.ai/code/artifact/e0214ec1-cb40-4d2f-a8ca-eb840127bb77 . No ablation applies to this card. Teardown: dev server terminated by PID (14563, found via lsof on port 3457 — never by process name); no background monitors were armed.",
  "open_questions": [],
  "out_of_scope_findings": []
}

---

Generated by Claude Code

---

Generated by Claude Code