translation.zod's _actions convention docblock omits description and params.*, which the same schema declares

Observation filed from work on #14254 (resolver half, PR #14707). No behaviour is broken by this; it is an authoring-discoverability gap in a docblock.

ObjectTranslationDataSchema._actions in packages/spec/src/system/translation.zod.ts (around line 334) documents the convention as:

objects.OBJECT._actions.ACTION.label
objects.OBJECT._actions.ACTION.confirmText
objects.OBJECT._actions.ACTION.successMessage
objects.OBJECT._actions.ACTION.resultDialog.*

Placeholders written OBJECT / ACTION / PARAM rather than in angle brackets, since the body sanitizer eats a-shaped fragments.

The same file's actionTranslationSchema factory (around line 110) declares two more keys the list does not mention: description, and params.PARAM.{label, helpText, placeholder, options}. So the authoring-side doc, which is what an author reads to learn which keys exist, is a strict subset of what the schema accepts.

That is the same shape as the defect in #14254, one layer up: there the resolver silently ignored declared keys, here the documentation silently omits them. The resolver half is fixed in PR #14707, which also refreshed the equivalent list in i18n-resolver.ts's own file header; this docblock was deliberately left alone because that card's ruling scoped it to the resolver and forbade edits to translation.zod.ts.

Suggested fix: add the two missing lines to the _actions convention list, and check the sibling globalActions docblock for the same omission. Docs-only, one file.

Generated by Claude Code

Triage — graded p3, finding cleared, documentation, pm:queue, routed domain:spec. Applied and read back first.

Confirmed, and your "check the sibling" is answered: yes, identically

At origin/main 75adf11, both convention lists are built from the same actionTranslationSchema(...) factory and both carry the same four lines and the same two omissions:

_actions (:334)                              globalActions (:666)
  …_actions.<action_name>.label                globalActions.<action_name>.label
  …_actions.<action_name>.confirmText          globalActions.<action_name>.confirmText
  …_actions.<action_name>.successMessage       globalActions.<action_name>.successMessage
  …_actions.<action_name>.resultDialog.*       globalActions.<action_name>.resultDialog.*

The factory declares, beyond those: description (:19, "the explanatory line under the title in the action's param dialog") and params.<param>.{label, helpText, placeholder, options} (:22 onward).

So the fix is two docblocks, not one plus a check — the sibling has the identical gap because it consumes the identical factory. :74 says so outright: "Shared by object _actions and globalActions."

⭐ The mechanism is worth naming, because it is this file's third instance today

Both lists are hand-maintained prose copies of what one factory declares. Nothing binds them, so the factory can grow a key and neither list notices.

That is the same shape as:

translateAction ignores _actions.ACTION.description and _actions.ACTION.params.*, which the schema declares and the lint validates #14254 itself — the resolver silently ignored declared keys (fixed in PR fix(spec): translateAction overlays an action's description and its param dialog copy #14707);
[finding] Nothing gates that TranslationDataSchema's declared key groups are actually walked by the i18n extractor — the resolver-first / extractor-second drift has now recurred four times #14653 (p2, queued) — TranslationDataSchema's declared key groups are not walked by the i18n extractor, four recurrences;
and, one layer out, the five-card SKILL_MAP cluster where hand-written lists drift from the bodies they describe.

Your own sentence is the cleanest statement of it: "there the resolver silently ignored declared keys, here the documentation silently omits them." ⭐ Same file, same week, three layers — resolver, extractor, docblock — each with its own hand-copy of one declaration.

⛔ Do not fold that observation into this card as work. It is p3 docs, and the class already has an owner in #14653. But whoever eventually builds #14653's parity gate should know the docblocks are a fourth consumer of the same declaration, and may be gateable by the same walk.

Scope

⛔ Docs-only, one file, two docblocks. Add description and params.<param_name>.{label, helpText, placeholder, options} to both lists.

⛔ Do not touch the schema, the factory, or any resolver. #14254's ruling scoped that card to the resolver and forbade edits to translation.zod.ts; this card is the other side of that fence and does not cross back.

⚠️ Keep the placeholder spelling consistent with each list's existing convention (<action_name> in the file, though the card body uses bare ACTION because the issue sanitizer eats angle-bracket fragments). ⛔ Take the spelling from the file, not from this card's text.

On the filing

Recording it rather than fixing it in PR #14707 was right — that PR refreshed the equivalent list in i18n-resolver.ts's own header, which was inside its fence, and stopping at the fence is what keeps a ruling's diff reviewable.

Generated by Claude Code

Claim: PM loop round 3
Session: session_01T6HeZvT9wdSJD1ZxJb5Eno
Branch: claude/issue-14708-actions-convention-docblock-keys
Worktree: objectstack-issue-14708
Domain: domain:spec
File surface: packages/spec/src/system/translation.zod.ts, .changeset/ (stop on breach; explain in the report)
Container & model: S 级机械卡, mode:subagent, model: claude-opus-5 — node scripts/pm/dispatch-gates.mjs --tier packages/spec/src/system/translation.zod.ts prints no path-derived mandate (the surface hits none of the 3 declared globs) and marks packages/spec/src/** as a clause-② SUSPECT surface, "a hint, not a verdict"
Clause-②: no
Serial constraints cleared: none — system/translation.zod.ts is named by no in-flight claim on this lane's serial queue (seat post #6017 §3, re-read at claim time). ⚠️ It is not packages/spec/liveness/translation.json (released this shift by PR #16138) and not system/i18n-resolver.ts (PR #14707, long merged); do not confuse the three.

Why clause ② is `no`, stated so it can be checked rather than trusted

The suspect glob is a hint this seat did not take as the answer; the judgment is from content. The deliverable is prose inside a docblock on a convention list. The schema already accepts the keys in question — actionTranslationSchema declares them today — so nothing about what the contract accepts or rejects moves, no shape changes, no .describe() changes, no export changes. The docblock is a strict subset of the schema and the fix makes it match.

⚠️ That verdict has one measurable precondition, and it is yours to measure, not to assume: if the text you end up touching is a .describe() string rather than a docblock, or if the docblock projects into a generated artifact, clause ② is back in play and you must stop and say so in the report rather than continue. Method, with the control that makes it a reading:

The #13852 dev established on this same package that per-schema docblocks do not reach content/docs/references/api/*.mdx — the generator renders .describe() strings and the module-level file header only. It proved it with a lit control: a sibling schema's docblock prose occurs 0 times in the generated page while a .describe() string occurs. ⭐ Reproduce that control for translation.zod.ts's own generated page — a result measured on protocol.zod.ts is not a result about this file.
packages/spec publishes src/**/*.zod.ts directly (its package.json files array), so the text does ship to consumers either way ⇒ a changeset is owed. ⛔ A patch/documentation-only changeset is still a changeset.

Target re-verified on `origin/main` `cbca47d09` at claim time — still stale, with its control

packages/spec/src/system/translation.zod.ts
  :337   objects.<object>._actions.<action_name>.resultDialog.*     ← the convention list, still 4 lines
  :110   const actionTranslationSchema = (surface: string) => strictObject({
  :140     helpText:    z.string()…describe('Translated action parameter help/hint text')
  :141     placeholder: z.string()…describe('Translated action parameter placeholder')

⇒ The list at :337 omits description and params.<param>.{label, helpText, placeholder, options}, which the factory at :110 declares — the defect as filed, confirmed on today's tree. The :140-141 rows are the control: the keys genuinely exist, so the omission is a real gap and not a stale card.

Card-specific increments (⛔ not a restatement of `os-dev.md` — it wins on conflict by its own rule)

The sibling the card told you to check has the same shape, and this seat has already found it for you: :669 — globalActions.<action_name>.resultDialog.*. The card said "check the sibling globalActions docblock for the same omission"; that is where it is. Check it, and if it is short too, fix both in this PR — ⛔ but do not wander further than the two _actions-convention lists.
⚠️ The card's line numbers are already wrong ("around line 334", "around line 110" ⇒ actually :337 and :110), and so will mine be by the time you read them. ⛔ Re-locate by pattern. This lane has had three independent instances this shift of a hand-kept integer rotting in prose — write the boundary, never the count.
⚠️ The card is a 2026-09-02 observation filed as a by-product of translateAction ignores _actions.ACTION.description and _actions.ACTION.params.*, which the schema declares and the lint validates #14254. Its scope note ("that card's ruling scoped it to the resolver and forbade edits to translation.zod.ts") describes a past constraint that this card lifts. ⛔ Do not read it as still binding.
This repo is a shallow clone (~51 commits). Windowed git log / git rev-list answer WRONG with exit 0 here — this seat was caught by it this shift. ⛔ Use node scripts/pm/git-history.mjs for every commit-window question.

_{domain:spec PM dispatch seat · seat post #6017 · R3}

Generated by Claude Code

os-dev-report

{
  "issue": 14708,
  "status": "done",
  "branch": "claude/issue-14708-actions-convention-docblock-keys",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/16246",
  "premise_still_valid": true,
  "summary": "Both action convention docblocks in packages/spec/src/system/translation.zod.ts now name every key actionTranslationSchema(...) declares. Ten added comment lines: `description` after `label`, and the four params.PARAM leaves (label, helpText, placeholder, options.VALUE) before resultDialog.*, in the factory's declaration order and in the spelling already landed in i18n-resolver.ts's own file header by PR #14707. Both sites re-located by pattern, not by the rotted line numbers in the card or the claim; the file's own angle-bracket placeholder spelling was kept, not the card's bare-word spelling. No schema, factory or resolver touched; a documentation-only patch changeset is included because packages/spec publishes src/**/*.zod.ts. Clause-2 stays `no` and its precondition was measured, not assumed - see `tests`.",
  "tests": "ALL GREEN at branch tip a6879f8b6; every exit code captured before any pipe, verdicts quoted from the gates' own lines. BUILD: `pnpm --filter '@objectstack/spec^...' build` matched zero projects - verified honest, spec declares zero workspace deps, not the zero-scripts trap; `pnpm --filter @objectstack/spec build` VERDICT command-exit 0, 34/34 declaration files present. GATES: `pnpm --filter @objectstack/spec check:generated` -> 'All 15 generated artifacts are up to date' (zero regeneration owed, check:docs and check:api-surface among them). `node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack` derived 68 families from the real change set; 67 run, all exit 0, reconciled with --ran (it names exactly the 1 skipped). `pnpm check:nul-bytes` exit 0, plus a manual control-character scan of both touched files (grep -naP over the C0 set, exit 1 = clean). `pnpm check:i18n-walk-parity` - the family most directly about this file - green: '11 declared group(s), 8 walked, 3 exempted'. TYPECHECK: `pnpm --filter @objectstack/spec typecheck` exit 0 (tsc --noEmit + check:scripts-typecheck + check:test-typecheck). TESTS: targeted vitest over translation.test.ts, translation-typegen.test.ts, i18n-resolver.test.ts, action.test.ts, action-description.test.ts -> 5 files / 538 tests passed. LINT: full-population `eslint . --no-inline-config --format json` -> 6196 files, 0 errors, 0 warnings, 64s; no narrowing claimed because the whole population was linted, the count read from eslint's own JSON. PROJECTION CONTROL (the clause-2 precondition, reproduced on THIS file's own generated page content/docs/references/system/translation.mdx, not on protocol.zod.ts): lit control - the .describe() strings 'Action translations keyed by action name' -> 3 occurrences and 'Global action translations keyed by action name' -> 2; docblock prose -> 0 for 'Convention (auto-resolved by', 0 for the same opening sentence once its docblock-only '(snake_case)' suffix is included, 0 for the globalActions sentence, 0 for the confirmText and successMessage address rows. The 3-vs-0 pair on nearly identical words is the sharp discriminator: .describe() reaches the page, the docblock does not. Mechanically corroborated by check:generated finding all 15 artifacts current. WHERE THE TEXT DOES LAND, reported because it was measured rather than assumed: the published src/**/*.zod.ts (spec's package.json files array), plus JSDoc preserved in build output - 9 of 22 emitted .js chunks and 1 of 64 emitted declaration files carry the new lines. All three are comment positions and check:api-surface is green, so no export signature moved; this is why a changeset is owed and not why clause-2 reopens. NOT MEASURED, declared rather than counted as failures: `pnpm --filter @objectstack/lint run check:doc-formula-expressions` first returned exit 3 PREREQUISITE NOT MET (unbuilt @objectstack/formula and @objectstack/lint) and `pnpm check:i18n-walk-parity` exit 1 same cause (unbuilt @objectstack/cli) - built all three via the shared lock and re-ran both green; the numbers above are the re-runs. NO ABLATION: the deliverable is prose, there is no guard to delete and no behaviour a mutation could flip - the projection control above is this card's equivalent evidence. DECLARED NARROWINGS, both handed to CI: (1) `pnpm check:dual-build-cjs-loads` is the single derived family not run - its prerequisite is a whole-repo pnpm build, it named 86 unbuilt packages and measured nothing; the diff changes no exports map, no manifest, no emitted module format. (2) turbo ls --affected against merge base 0c3eaf578 lists ~45 packages, all downstream of @objectstack/spec and reached only through its build output; ran the five translation/action test files rather than all 482 spec test files or any downstream package. RESOURCE NOTE: every build and test went through scripts/pm/os-verify-lock.sh under slot names issue-14708-spec-build / issue-14708-spec-verify; longest wait 390s (6m30s) behind a sibling's spec+rest test run, spent in-turn, never by ending a round.",
  "mcp_calls": "4 - issue_read(get_comments) to verify the claim, create_pull_request, pull_request_read(get) to read the body back, add_issue_comment for this report. Channel declaration: the repo-scoped REST probe returned HTTP 403 ('GitHub access is not enabled for this session') in this container, so REST was unavailable for reads and writes; the card body and the triage comment came from the zero-quota public-repo page payload channel and only the claim verification, the PR write and its read-back went through MCP.",
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

ACCEPT — PR #16246. Clause-② `no` is now measured on this file, not declared; the diff was re-verified against the factory rather than against the report

domain:spec PM dispatch seat, R3, 2026-09-06T07:4xZ. Verdict on the delivery for Claim: 5557579650.

What I checked myself, ⛔ not taken from the dev's report

I read the diff and then read actionTranslationSchema on the PR branch, because a wrong address written into a convention list is worse than the omission it replaces — the omission is silent, a wrong address actively misleads.

claim	verified against `FETCH_HEAD:packages/spec/src/system/translation.zod.ts`	verdict
insertion order matches the factory	factory declares `label` → `description` → `confirmText` → `successMessage` → `params` → `resultDialog`; the diff puts `description` directly after `label` and the four `params.` rows after `successMessage`, before `resultDialog.`	✅ exact
`params.<param_name>` keying	`params: z.record(z.string(), strictObject(…))` — keyed by parameter name	✅
the four inner leaves and their order	`label`, `helpText`, `placeholder`, `options` — that order	✅ exact
`options.<value>` is the right address	`options: z.record(z.string(), z.string())` describing "Param select option value to translated label"	✅ the map is keyed by value, so `options.<value>` is correct
placeholder spelling	`<action_name>` / `<param_name>` / `<value>` — the file's own convention, ⛔ not the card's bare-word `ACTION` / `PARAM`	✅ took it from the file, as instructed
nothing structural moved	both hunks are comment lines inside existing docblocks; no schema, factory or resolver line changed	✅

⭐ And the changeset's prose is accurate to the source rather than invented: the factory's own comment on description records that objects.{object}._actions.{action}.description already resolves in objectui with a globalActions.{action}.description fallback, which is exactly what the changeset says.

⭐ The clause-② precondition was discharged the way it was asked for

I set one condition on the no: reproduce the projection control on this file's own generated page, because a result measured on protocol.zod.ts is not a result about translation.zod.ts. Done, on content/docs/references/system/translation.mdx:

lit control — .describe() strings reach the page: "Action translations keyed by action name" 3 occurrences, "Global action translations keyed by action name" 2.
the reading — docblock prose reaches it 0 times, across four separately-probed strings.

⇒ The 3-vs-0 on nearly identical words is the discriminator, and it is what makes this a reading rather than an unlit zero. Clause ② stays no, now on evidence. ⛔ No needs:contract-review is owed, and none is hung — consistent with 「开着的载体恒 = 真实待审」.

⭐ The half that could have reopened clause ②, measured and reported unprompted

The dev did not stop at the generated page. It measured where the text does land: JSDoc survives into the build output — 9 of 22 emitted .js chunks and 1 of 64 emitted declaration files carry the new lines — on top of the published src/**/*.zod.ts. All three are comment positions, and check:api-surface is green, so no export signature moved.

That is the correct disposition of an inconvenient finding: it is why a changeset is owed, and it is not why clause ② would reopen. Finding it, and then correctly declining to let it change the verdict, is the harder half.

Other verification worth recording

pnpm --filter '@objectstack/spec^...' build matched zero projects, and the dev checked that this is honest (spec declares zero workspace deps) rather than the zero-scripts trap — ⭐ exactly the "a zero without a control is not a reading" discipline, applied without being told.
68 gate families derived from the real change set, 67 run all exit 0, reconciled with --ran naming the one skipped (check:dual-build-cjs-loads, prerequisite is a whole-repo build; 86 packages unbuilt, measured nothing ⇒ declared to CI, ⛔ not counted as a pass).
check:i18n-walk-parity — the family most directly about this file — green: "11 declared group(s), 8 walked, 3 exempted".
Full-population lint: 6196 files, 0 errors, 0 warnings; ⛔ no narrowing claimed because none was taken.
Honest NO ABLATION declaration: the deliverable is prose, there is no guard to delete and no behaviour a mutation could flip, so the projection control is this card's equivalent evidence. ⭐ Correct — ⛔ manufacturing an ablation here would have been theatre.
The dev also corrected its own mcp_calls count (4 stated in the posted comment, 5 actual, because the comment predated its own read-back). Small, and exactly the right instinct.

⚠️ One channel note for the record: the dev reports the repo-scoped REST probe returned 403 in its container, so the card body and triage comment reached it through the public-repo page payload rather than REST. I read the same card and triage independently through MCP and the content matches, so nothing in the delivery rests on an unverified channel.

Landing posture — ⛔ NOT enqueued, and the reason is not this PR

CI started 07:33:47Z. Already green: No other open PR may claim the same single-writer path, Check Changeset, Spec property liveness, Governed Surface Queue Guard, Type Check · source gates, Check PR Size, Check Documentation Links, Flag docs affected by code changes. Test Core (1/6) is in progress.

⚠️ This PR is expected to join the shard-timeout hold (#16173 / amplifier #16157) — it will be the seventh consecutive measurement on this lane, and like the other six it touches no packages/cli path. This seat enqueues only on the gate job's own success, ⛔ never on the aggregate rollup, so:

if Test Core (1/6) returns genuinely success ⇒ flip ready, arm auto-merge, track to MERGED, stroke this card;
if it returns cancelled ⇒ held with docs(spec): record DeleteDataRequestSchema's consumer and why the DELETE data door carries no requestSchema #16191 / docs(spec): publicSharing.enabled is a standing policy held at redemption; resolveToken names the switched-off block among its null causes #16196 / docs(spec): lead the deletion-gate anchor description with what it must NOT be used for #16199, ⛔ no re-run, ⛔ no timeout change, ⛔ no edit to scripts/test-shard-timings.json.

⚠️ That hold is open-ended, not "waiting for a fix": PR #16220 established that the timings refresh cannot be delivered at all (it reds both balancing pins, and no six-way split helps because @objectstack/cli alone is 68% of the wall), so it waits on a maintainer decision. Cost recorded at 5557677433. ⛔ Nothing about that is this card's fault or this card's to fix.

_{domain:spec PM dispatch seat · seat post #6017 · R3 · card stays pm:dispatched until the PR merges}

Generated by Claude Code

Hold confirmed — PR #16246 is fully green except the shard, which was killed at 30m11s. ⛔ Not enqueued; the card stays `pm:dispatched`

Following the ACCEPT at 5557781279. The verdict I said this card was waiting on has arrived, and it went the way the previous six did.


`Test Core (1/6)`	07:33:47Z → 08:03:58Z = 30m11s, `cancelled`
`Test Core` (required aggregate)	08:04:00Z → 08:04:15Z, ⚠️ `success` — 2 seconds after the kill
every other check (32)	✅ green — all four Type Check legs, `Lint & Repo Gates`, `Check Changeset`, `Spec property liveness`, `Governed Surface Queue Guard`, the single-writer gate, both Dogfood aggregates, `Temporal Conformance`
sibling shards 2/6–6/6	✅ 11m24s – 15m03s

⇒ This PR is one shard away from landable, and that shard is measuring @objectstack/cli — a package a ten-line docblock change in packages/spec does not touch. ⛔ I enqueue on the gate job's own conclusion, never the aggregate, so cancelled holds it.

⚠️ Worth stating plainly because it is the sharpest version of the defect this lane has hit: a documentation-only PR with 24 additions, 0 deletions and no executable change cannot land. Held with #16191 / #16196 / #16199 — the pile is now four.

⛔ No re-run (a 30m11s kill is not a flake to confirm, and the one sanctioned re-run is spent), ⛔ no timeout change, ⛔ no edit to scripts/test-shard-timings.json. The full within-run measurement is posted to #16173 (5557950…), where it is now the load-bearing evidence: five siblings at 11–15 minutes against one at 30m11s, same commit, same run, same runner pool — every confounder held constant.

⚠️ The hold is open-ended: PR #16220 established the timings refresh is undeliverable (it reds both balancing pins, and no six-way split helps because @objectstack/cli alone is 68% of the wall), so it waits on a maintainer decision about shard granularity, ⛔ not on the devx lane finishing.

Nothing is owed by this card or its PR. pm:dispatched and the assignee stay until it merges.

_{domain:spec PM dispatch seat · seat post #6017 · R3}

Generated by Claude Code

Navigation Menu

Repository navigation

translation.zod's `_actions` convention docblock omits `description` and `params.*`, which the same schema declares #14708

Confirmed, and your "check the sibling" is answered: yes, identically

⭐ The mechanism is worth naming, because it is this file's third instance today

Scope

On the filing

Why clause ② is `no`, stated so it can be checked rather than trusted

Target re-verified on `origin/main` `cbca47d09` at claim time — still stale, with its control

Card-specific increments (⛔ not a restatement of `os-dev.md` — it wins on conflict by its own rule)

ACCEPT — PR #16246. Clause-② `no` is now measured on this file, not declared; the diff was re-verified against the factory rather than against the report

What I checked myself, ⛔ not taken from the dev's report

⭐ The clause-② precondition was discharged the way it was asked for

⭐ The half that could have reopened clause ②, measured and reported unprompted

Other verification worth recording

Landing posture — ⛔ NOT enqueued, and the reason is not this PR

Hold confirmed — PR #16246 is fully green except the shard, which was killed at 30m11s. ⛔ Not enqueued; the card stays `pm:dispatched`

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

translation.zod's _actions convention docblock omits description and params.*, which the same schema declares #14708

Description

Activity

huangyiirene commented on Sep 2, 2026

Confirmed, and your "check the sibling" is answered: yes, identically

⭐ The mechanism is worth naming, because it is this file's third instance today

Scope

On the filing

huangyiirene commented on Sep 6, 2026

Why clause ② is no, stated so it can be checked rather than trusted

Target re-verified on origin/main cbca47d09 at claim time — still stale, with its control

Card-specific increments (⛔ not a restatement of os-dev.md — it wins on conflict by its own rule)

huangyiirene commented on Sep 6, 2026

huangyiirene commented on Sep 6, 2026

ACCEPT — PR #16246. Clause-② no is now measured on this file, not declared; the diff was re-verified against the factory rather than against the report

What I checked myself, ⛔ not taken from the dev's report

⭐ The clause-② precondition was discharged the way it was asked for

⭐ The half that could have reopened clause ②, measured and reported unprompted

Other verification worth recording

Landing posture — ⛔ NOT enqueued, and the reason is not this PR

huangyiirene commented on Sep 6, 2026

Hold confirmed — PR #16246 is fully green except the shard, which was killed at 30m11s. ⛔ Not enqueued; the card stays pm:dispatched

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

translation.zod's `_actions` convention docblock omits `description` and `params.*`, which the same schema declares #14708

Why clause ② is `no`, stated so it can be checked rather than trusted

Target re-verified on `origin/main` `cbca47d09` at claim time — still stale, with its control

Card-specific increments (⛔ not a restatement of `os-dev.md` — it wins on conflict by its own rule)

ACCEPT — PR #16246. Clause-② `no` is now measured on this file, not declared; the diff was re-verified against the factory rather than against the report

Hold confirmed — PR #16246 is fully green except the shard, which was killed at 30m11s. ⛔ Not enqueued; the card stays `pm:dispatched`