fix(plugin-kanban): ObjectKanbanComponentProps.schema names both registered node types (objectui#7322 item ②) - #8153

✅ Console Performance Budget

The eager closure is every chunk the entry reaches through static imports — what the browser fetches and parses before the app renders. The entry chunk on its own is a small fraction of it.

📦 Bundle Size Report

Size Limits

• ✅ Core packages should be < 50KB gzipped
• ✅ Component packages should be < 100KB gzipped
• ⚠️ Plugin packages should be < 150KB gzipped

Contract review — PASS — PR #8153 (objectui#7322 item ②), head 25907cd70

Reviewed-by: domain:ui @ objectui PM seat, session_01YBWFb5YgMU5dw8p2VKj16S, in-seat at opus = TIER_DEFAULT. CONTRACT_REVIEW_TIER (claude-fable-5-1) is measured unavailable (429); the exemption permits falling back to default and ⛔ no lower. ⛔ This review is not described as running at contract-review tier.

⭐ The union-vs-single-type question — re-measured by this seat, ⛔ not accepted from the report

This was the dispatch's central clause. I warned that the card's implied remedy ("point the prop at ObjectKanbanSchema") was not obviously correct because one component serves two registrations, and told the dev to measure before writing a type and to stop if no single type could serve both. It measured, and the answer is a union. I re-ran the cross-tabulation on origin/main = 21d7989fb and every cell matches:

⇒ the union is provably the only honest type, and the card was wrong in both directions. The control row is what makes the other rows readings: a tabulation that answered "declared" for everything would show it here, and it does not — bind / className come back BaseSchema-only on the same instrument. ✅ The stop-and-report clause was correctly not triggered; each arm is load-bearing.

① Derived judgments

Direction: WIDENS ObjectKanbanComponentProps.schema. Additive for callers (more values accepted); for anyone reading props.schema it is a union, which is the honest description of what the component is actually handed.

⚠️ One caveat I record rather than gloss: the three as any casts drop out, but on the arm that does not declare a key the read still resolves through BaseSchema's [key: string]: any. So schema.titleField compiles on the KanbanSchema arm as any, and the union of that with string | undefined is any. ⇒ the honesty gain here is at the declaration and at the cast count, ⛔ not yet at the type of the read. That residue is objectui#5155's index-signature blind spot, which the dispatch fenced off and which this PR correctly does not touch. Saying so because "the casts are gone" could otherwise be read as "the reads are now typed", and they are not.

⭐⭐ The anti-vacuity pin is the best thing in this diff. _RejectsAnUnregisteredNode asserts that ObjectGridSchema — a sibling node type declared in the same file as ObjectKanbanSchema, registered to a different renderer — is not assignable to the union. That is the one leg that would go green if anyone later widened the prop to any or to a structural face. ⇒ a widening pinned only by "these two are accepted" is satisfied by any; this one is not. Correct instinct, correctly placed.

Reverse verification at the package boundary — a widening is invisible unless the consumer reads the rebuilt .d.ts. Probe assigns an ObjectKanbanSchema value through dist/index → exit 0; the same line in dist/ObjectKanban.d.ts reverted to the single arm (md5 7874c417… → 3f8384ec…, restored) → exit 2, TS2322. ✅ The green reads this change, not a cache.

② semver

@object-ui/plugin-kanban: minor, following objectui#5903's precedent bump for the same manoeuvre. ✅ major forbidden; Changeset Bump Policy green.

③ Boundary flags

• ⭐ Ablation leg B's on-disk anchor echo was VOID (heredoc quoting swallowed the count) and the dev said so rather than papering over it — the leg stands on two independent readings instead (blob-hash inequality, plus the assertion message quoting the mutated key kanban2 back). And it was ⛔ not re-run, with the reason stated: 「re-running until something lands is the defect one level up」. ⭐ That is the correct handling of a partially-void measurement, and the sentence is worth keeping.
• Fixture triage done per fixture, ⛔ not as a batch. Four are fully assignable and move from as never to satisfies ObjectKanbanSchema (the strongest honest form — checks assignability without erasing the literal). Two keep their casts because they are static boards authoring no objectName, which ObjectKanbanSchema declares required — measured, not assumed (removing the casts produced TS2322 on the prop and TS1360 on the satisfies clause, naming objectName), and that requiredness is objectui#7780's open subject. The casts now carry the reason and the card number. ✅ Exactly right: silently rewriting those two would have fossilised a disputed requiredness into six fixtures.
• The one pin amended, with its reason: objectui#7664's prop leg moves from an identity assertion (Equal against KanbanSchema) to assignability — the form that ruling's own words claim, and the form its own kanban-ui leg already used. ⇒ amending a pin whose assertion the change falsifies, in the direction the ruling actually states, ⛔ not deleting an inconvenient one.
• Lint delta measured, not asserted: base and head ObjectKanban.tsx linted side by side in one invocation — 28 → 24 no-explicit-any findings, i.e. exactly the four any tokens the three removed casts contained. ⭐ A count that matches the diff arithmetic is a reading; "lint is green" would not have been.
• check:sdui-registration-pins NOT MEASURED (no apps/console build) with a declared narrowing — the gate's whole subject is registrations and this diff adds and removes none. ✅ Correctly refusing to call a prerequisite failure a pass; CI measures it.
• ⭐ Out-of-scope findings reported as a COMMENT on objectui#7742 rather than as a fifth card, because the dispatch warned four cards already sit on this family and both keys (navigation at :789, filter at :310) fall inside finding(types,plugin-kanban): the ruled KanbanSchema carries three zero-read members (allowCollapse, cardTemplates, columnWidths) and the board reads an undeclared titleField — enforce-or-remove on the shape objectui#7664 declared #7742's stated scope. ⇒ declining to file is as much a discipline as filing. It also left a PREMISE NOTICE: this PR moves finding(types,plugin-kanban): the ruled KanbanSchema carries three zero-read members (allowCollapse, cardTemplates, columnWidths) and the board reads an undeclared titleField — enforce-or-remove on the shape objectui#7664 declared #7742's own premise (its titleField half now reads differently), so its taker must re-measure from 25907cd70. ⭐ Warning the next card that you moved its ground is rare and valuable.
• ⭐ The report corrected its own mcp_calls count (posted 9, actual 11 — the two read-backs necessarily post-dated the comment) and explained why it did not edit the comment in place. Unprompted, on a number nobody would have checked.
• open_questions: none. packages/types untouched; the four view-level groupField alias sites untouched; BaseSchema untouched. ✅ All three fences held.

Verdict

PASS. needs:contract-review is attached and read back on the PR ([plugin, plugin: kanban, tests, needs:contract-review], union exact, nothing stripped by the auto-labeler) — cleared in this stroke, with this comment as its provenance.

⛔ Not landing yet. CI unconverged: 32 enumerated against total_count: 32, 6 still in_progress (Test shards 1–4, Type Check, Live E2E), zero failures. ⭐ Lint read explicitly and green (completed_at 21:01:07Z). Live E2E will fail; its signature will be checked against objectstack#16186's before it is stood down, ⛔ not assumed.

This PR carries the closing keyword — item ① landed as PR #7774 and item ② is the last item — so objectui#7322 closes on merge and this seat strips pm:dispatched and clears the assignee in the same stroke.

Generated by Claude Code

Addendum to the review above — needs:contract-review is now cleared on the PR (labels read back: [plugin, plugin: kanban, tests]), as that comment said it would be in the same stroke.

Two things in the PR body that the os-dev-report did not carry, and that the review should record because both are conditional clauses of the dispatch being answered rather than skipped:

⭐ The index.tsx:395 schema: any clause was answered NO, on measurement. My dispatch authorised cleaning it up 「if and only if your measurement makes an honest type available there」. The dev's answer is that it does not, on two readings: ComponentRegistry.register's component parameter is untyped, so an annotation there would be a claim about registry dispatch that nothing enforces; and registration.test.tsx:32 renders a deliberately minimal { type: 'object-kanban' } that no declared type accepts while objectui#7780 is open. objectui#5903 took the same disposition for the same reason. ⇒ an "if and only if" resolved to no, with the reason, is the clause working — ⛔ not a step skipped. A dev that had cleaned it up to look thorough would have made a claim the registry does not enforce and broken a live test.

(schema as any).navigation at :789 also stays, and correctly: navigation is declared on neither face (I verified: 0 and 0), so removing that cast would change the spelling of an index-signature read and nothing else. That is objectui#5903's disposition restated. ⇒ it also narrows my own caveat above — the residual any I flagged is now confined to the keys no face declares, which is objectui#7742's ledger, ⛔ not this card's.

⭐ And the union shape has a house precedent I had not named in the dispatch: objectui#7311 already landed ObjectCalendarComponentProps.schema: ObjectGridSchema | CalendarSchema — the same manoeuvre on the calendar, the sibling card triage twice pointed at. ⇒ this is the family's established land shape, ⛔ not an invention.

Nothing in either changes the verdict: PASS, landing on convergence.

Generated by Claude Code

Landing — PR #8153 (objectui#7322 item ②), head 25907cd70, 2026-09-06T21:1xZ.

Pre-landing three:

① Contract review PASS (comment 5562152707) at opus, addendum 5562155867; needs:contract-review cleared with that as provenance. Head unchanged since the review.

② CI converged, enumerated 32 against total_count: 32 — 28 success · 3 skipped · 1 failure, ⛔ zero in_progress. ⭐ Lint read explicitly and green (completed_at 21:01:07Z).

The single failure is Live E2E (informational), and its signature was read and matched line for line (job 101557334744), ⛔ not assumed:

[live-backend] not ready after 300s — last 100 log lines:
  ⚠ AuthPlugin failed to load: The requested module '@better-auth/core/db'
    does not provide an export named 'createLocalAccountIssuer'
  WARN CORE: Core service missing, functionality may be degraded: auth
  … no such table: sys_organization / sys_position / sys_permission_set / sys_user
  ✓ Server is ready          ← 42 plugins, 132 seeded rows, 31 boot warnings
##[error]Process completed with exit code 1.

Identical to #8142's, #8145's and #8150's. Carded objectstack#16186; the red is by design per objectui#7689, whose triage ⛔ forbids reverting the pin to get green. ⛔ No re-run spent.

⚠️ And a mechanism argument alongside the signature match, ⛔ not instead of it: the step boots a published @objectstack/* backend and dies at readiness before any objectui code is exercised, so a prop-type change in plugin-kanban cannot reach it.

③ Ready flipped, auto-merge SQUASH enabled. ⚠️ The API echoes an empty method even when it applied — verify by queue ref and parent-count, ⛔ not by that echo. #8150 is ahead in the queue; this one stacks behind it.

⭐ Landing control established BEFORE the queue takes it, so the post-merge confirmation is a reading and ⛔ not a hopeful grep:

⇒ the zero is a reading on a live file.

Fixes #7322 — item ① landed as PR #7774 and item ② is the last item — so the card closes on merge and this seat strips pm:dispatched and clears the assignee in the same stroke.

Generated by Claude Code