✅ 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

契约复核 — 裁决:REFUSE(逐字采纳)

归属:domain:ui 派发席位,依 2026-08-31 裁定席内完成;董事席链条为可选事后审计,非放行必要条件;⛔ 无「等外部席位」的无限期状态。

档位与核验:隔离 fable 子代理(⛔ 契约复核不适用额度耗尽豁免降档)。transcript 中 harness 逐消息盖章的 model 字段 76 次读数全部 claude-fable-5-1,0 次回退。亮控:同一 grep 在本会话其它 transcript 上读出 claude-opus-5(30 份)与 <synthetic>(1 份)—— 该 grep 能读出非 fable 值,所以这里的零是读数。达标 ⇒ 逐字采纳。父会话只有两个合法动作:逐字采纳或整体作废;⛔ 不修剪、不润色、不挑着采纳。

独立性:只喂卡片 #7645、既有裁定与 PR 本身;⛔ 未喂派发简报、⛔ 未喂本席结论;简报按对抗性写法。

⛔ needs:contract-review 两个载体都不清除。 REFUSE 就是 REFUSE。补救是纯文字、限于 37 行的 changeset 一个文件,代码与 pin 复核判为 sound。补救推上去后重跑契约复核,通过再清标。

一处已知渲染损耗:下文有三处尖括号泛型(Equal<…>、Extract<…>、箭头函数类型),GitHub 正文净化器会吞掉尖括号内容。原文以本地 transcript 为准;⛔ 此处不改字,只作说明。

Contract review — objectui PR #7662 (card #7645)

Head 41ff58ff, base f7cf7e8a, merge-base with current main 2ce2612d; no drift on any touched type file; merge-tree clean. 32 CI checks: all success or skipped. Everything below was measured on a detached worktree at the head sha plus a second at the merge-base, with my own tsc probe battery (base leg exit 0, head leg exit 0, and a head-probe-on-base cross-control that fired on exactly the six predicted lines — the instrument is lit).

① Derived judgments

1. SchemaRegistry['kanban'] value: DeclarativeKanbanSchema → BaseSchema & { type: 'kanban' } (packages/types/src/registry.ts; external view confirmed in built dist/registry.d.ts:86). RIGHT as a type decision: A measured unreachable (phantom-deps gate doc line 127 says type-only imports are STRICT; head tree exit 0 with its verdict line; types has 0 workspace deps while plugin-kanban declares types), and both faces satisfy the new value (my P1/P2 pass against dist).
2. Accept-set as an assignment target: widened. { type: 'kanban' } alone is rejected at base (TS2741, missing columns) and accepted at head; { type:'kanban', draggable, objectName, garbage: 42 } accepted at head; { type: 'grid' } still rejected. Named in the PR body ("the value stops asserting") — RIGHT. Not named: that BaseSchema's [key: string]: any (base.ts:440, inside BaseSchema, closes at 441) is what admits arbitrary keys.
3. Read side: SchemaRegistry['kanban']['columns'] goes from DeclarativeKanbanColumn[] to any (H4: Equal<typeof cols, any> false at base, true at head). NOT NAMED by PR body or changeset. "Gets the base node shape instead" is literally true, but the consumer-visible effect is a silent type-safety loss, not an error. WRONG by omission in the migration note.
4. Assignability SchemaRegistry['kanban'] → DeclarativeKanbanSchema: compiles at base, TS2741 at head (H6). This is a compile break for any external consumer that flowed the registry value into the declarative face. The changeset's "this is a correction rather than a break" is WRONG.
5. SchemaRegistry['kanban'] ↔ plugin KanbanSchema: mutually assignable at head. My prediction that the entry would not flow into the plugin face MISSED (unused @ts-expect-error at pk-probe.ts:15) — the plugin face is all-optional except type, so the index signature satisfies it. Not a defect; it does mean the plugin-side pin proves nothing beyond the tag.
6. keyof SchemaRegistry / published ComponentType: byte-identical. 69 keys base and head, 'kanban' present; textual key-list diff identical, with a control that fires when kanban is removed; Extract<ComponentType,'kanban'> = 'kanban' on both trees. No silent narrowing — RIGHT.
7. Exported symbols: 0 export statements changed (the two grep hits for "export" are prose: changeset line 40, registry comment line 287). DeclarativeKanbanSchema still exported (index.ts:294) and still the ComplexSchema arm (complex.ts:1212). RIGHT.
8. Two new test files under src/__tests__/: not published (tsconfig.json excludes tests from dist; tsconfig.test.json is emit-free). RIGHT — no surface. Changeset-presence gate is presence-only, not per-package (script line 408), and plugin-kanban is in the same fixed group.
9. The PR's own pins are lit and in-program: tsc -p packages/types/tsconfig.test.json exit 0, pin file and registry.ts each once in --listFiles; the plugin pin run in isolation through the plugin's own tsconfig.test.json exit 0, resolving @object-ui/types to dist (1) not src (0). The whole-package plugin run is a precondition failure (45× TS2307 for unbuilt siblings), not a reading. Note the "Leg 3" invariant is redundantly held by my own H1/B1.
10. Latency claim re-measured: non-test SchemaRegistry\s*\[ = 0 at base and head; 6 at head are the PR's own pins; control ComponentPropsMap[ = 25. No @object-ui/types ComponentType consumer outside its own declaration (control: 136 React ComponentType hits). No SchemaRegistry import from @object-ui/types anywhere in the objectstack sibling (its 30 hits are objectql's own class). RIGHT.
11. A3 (ObjectKanbanProps) numbers reproduce exactly: 13 keys, shares 11, cannot express the 8 named, adds filter/titleField. RIGHT. (My first reading of this was a dark instrument — regex slid into ObjectCalendarPropsSchema — discarded and rebuilt.)
12. Fences honoured: plugin-kanban/src/types.ts untouched; no content/docs/releases/; 4 files in the diff.

② Semver grading

Declared: '@object-ui/types': minor (.changeset/7645-schema-registry-kanban-honesty.md, line 2).

Level: correct. AGENTS.md §版本号策略 line 239 is the standing rule: "changeset 里不要声明 major … objectui 自身的破坏性变更也标 minor（在正文里写清 breaking 语义即可）", enforced by scripts/check-changeset-no-major.mjs. So "needs a major bump" is the wrong conclusion in this repo even though the change is breaking (item 4). The rule's second half — the body must spell out the breaking semantics — is where the changeset fails:

• It gives a migration ("should name the face it actually means") but then states "which is why this is a correction rather than a break" — contradicted by H6 (loud compile break) and H4 (silent any).
• It omits the index-signature mechanism that makes member reads collapse to any without an error.
• Measured false: "19 members against the declarative face's own set, sharing only the type tag." Declarative own members are 7, plugin 19; shared by name: type, columns, onCardMove; onCardMove is (string, string, string, number) => void on both sides (only parameter names differ); columns is on both (required DeclarativeKanbanColumn[] vs optional KanbanColumn[], element types sharing 5 of 6 members).

③ Boundary-flag disposition

• Flag 1 (extend the family rule to every key?): ruled by PM — B now, C as follow-up; C exists as SchemaRegistry's docblock promises more than the layer can deliver — it owns the key set, not the values #7665 (2026-09-04T16:02:00Z, pm:queue). Answered.
• Flag 2 (validator and renderer honour different faces): ruled B then C; filed as finding(types/plugin-kanban): an authored kanban document can PASS safeValidateSchema and still render empty — the validator and the renderer honour different faces #7664 (needs-user-decision, 16:02:00Z). Escalated correctly.
• Declared not-measured check:sdui-registration-pins: accepted — the diff changes no registration; CI's Governed Surface Queue Guard passed regardless.
• My own miss (item 5) recorded above; it does not change the verdict.

Verdict

VERDICT: REFUSE — the code and pins are sound and the bump level is correct, but the changeset (the PR's published release-note input) contains one measured-false factual claim ("sharing only the type tag") and mischaracterises a measured compile break for external indexed-access consumers as "a correction rather than a break", while omitting that BaseSchema's index signature turns member reads into silent any. Remediation is confined to that 37-line file: name it a break for consumers who indexed SchemaRegistry['kanban'], state the any-read effect, and correct the shared-member sentence.

Implemented-by: claude/issue-7645-schema-registry-kanban-honesty
Reviewed-by: isolated subagent (host session_01KbJQ1y1J12nZxYzFWhP8Q3)

Generated by Claude Code

✅ 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 re-review (round 2) — claude-fable-5-1, adopted verbatim

Isolated review subagent. Tier verified from its transcript: 56/56 assistant turns stamped claude-fable-5-1, 0 fallback notices. Lit control: the same reader returns claude-opus-5 5810 times across this session's 102 transcripts, so the zero is a reading. Adopted verbatim — not edited, trimmed or paraphrased.

Contract review — objectui PR #7662 (card #7645), head 3748de9af

Reviewer: independent, this session (session_01KbJQ1y1J12nZxYzFWhP8Q3 host; no GitHub writes made, no shared checkout edited). Base for every measurement: merge-base of the PR head against current origin/main (a3eb5d07a) = 2ce2612df; PR commits since it: 41ff58ff7 (code) and 3748de9af (remediation). Both sides built and measured on detached scratch worktrees fetched into private refs (refs/review/7662-*), since removed; the primary checkout /home/user/objectui was read-only throughout (0 modified files, HEAD unchanged at 77b3cf1f5). Toolchain: node 22.22.2, pnpm 10.31.0, tsc 6.0.2; pnpm-lock.yaml, root and both package package.json/tsconfig.json blob-identical on both trees, so the two dist builds are of one dependency graph.

Instrument. One probe file (packages/plugin-kanban/src/__review7662__/probe.ts, identical bytes on both trees) compiled through the plugin's own tsconfig.test.json settings (paths: {}), so @object-ui/types resolves to the built dist (--listFiles: packages/types/dist/index.d.ts 1, packages/types/src/ 0, plugin-kanban/src/types.ts 1, both trees). Predictions for all 22 markers were written into the file before either run. Lit controls, all four fired on both trees: L1 Equal<DeclarativeKanbanSchema['draggable'], string> (TS2344), L2 GridSchema into DeclarativeKanbanSchema (TS2322), L3 IsAny<DeclarativeKanbanSchema['columns']> (TS2344), L4 { type: 'grid' } into SchemaRegistry['kanban'] (TS2322). Four non-vacuity pins (IsAny false for SchemaRegistry, KanbanSchema, DeclarativeKanbanSchema, the entry) passed on both. Totals: base exit 2 / 11 errors, head exit 2 / 7 errors — exactly as predicted, no marker missed. Every zero below is a reading from an instrument that produced non-zero on the same run.

① Derived judgments

1. SchemaRegistry['kanban']: DeclarativeKanbanSchema → BaseSchema & { type: 'kanban' }. Published view: dist/registry.d.ts:85 → :86-88; a diff -rq of the whole @object-ui/types dist between trees names registry.d.ts as the only differing file (plus the import lines that move with it). RIGHT as a type decision. A1 re-measured by ablation, not read from the gate's docs: check:phantom-deps exit 0 clean (✅ Every in-scope import is declared…), exit 1 with one type-only import added to registry.ts (packages/types/src/registry.ts:1:1 [undeclared-runtime] @object-ui/types imports '@object-ui/plugin-kanban' … (type-only)), blob restored to HEAD and git diff HEAD -- packages/ empty. A2 by my own manifest read: @object-ui/types 0 workspace deps, @object-ui/plugin-kanban 7 including @object-ui/types, so the edge is a cycle. Both faces satisfy the new value (M3 declarative→entry, M15 KanbanSchema extends entry: pass at head).
2. Accept set as an assignment target: widened. { type: 'kanban' } TS2741 at base, accepted at head (M11); { type: 'kanban', garbage: 42 } TS2741 at base, accepted at head (M13); { type: 'grid' } TS2322 on both (L4). The changeset names the mechanism ("the entry lacks the required columns member") — RIGHT.
3. Assignability entry → DeclarativeKanbanSchema: compiles at base, TS2741 at head (M2: Property 'columns' is missing in type 'BaseSchema & { type: "kanban"; }'). Changeset now calls it "Breaking, loudly, for consumers who indexed SchemaRegistry['kanban']" — RIGHT. (The prior REFUSE's item 4, remediated.)
4. Member reads collapse to any through BaseSchema's index signature (base.ts:440). entry['columns']: DeclarativeKanbanColumn[] at base, any at head (M6a/M6b, each the other's cross-control, exactly one fires per tree); entry['draggable'] likewise (M7); undeclared keys any on both trees (M8); entry['className'] stays string | undefined on both (M17, BaseSchema declares it). Changeset states the mechanism, the columns before/after, and the undeclared-keys caveat — RIGHT. (Prior item 3, remediated.)
5. Plugin face ↔ entry. KanbanSchema → entry: TS2322 at base, compiles at head (M4); entry → KanbanSchema: compiles on both (M5). So the plugin-side pin proves the tag and nothing narrower; neither changeset nor body claims more — RIGHT.
6. keyof SchemaRegistry / published ComponentType: unchanged. 69 keys on both commits (source via git show, and 69 in each dist/registry.d.ts), lists identical; control: dropping 'kanban' from the head list → 1 differing line. M9 Extract<ComponentType,'kanban'> = 'kanban' on both. RIGHT.
7. Exported symbol set of dist/index.d.ts: IDENTICAL between trees; DeclarativeKanbanSchema still exported (1). The two new test files are excluded from dist (**/__tests__/** in both packages' tsconfig.json; no dist/__tests__). No surface added — RIGHT.
8. Changeset member census (7 / 19 / shared 3 = type, columns, onCardMove; columns 6 / 6 sharing 5; onCardMove type-identical). My AST census through the TypeScript compiler API reproduces every number; controls: DeclarativeKanbanCard 9, self-intersection 7, disjoint pair (DeclarativeKanbanSchema ∩ KanbanColumn) 0, nonexistent interface null; M16 confirms onCardMove Equal on both trees. RIGHT. (Prior item "sharing only the type tag", remediated.) Precision note, not a defect: "the declarative draggable / onCardClick have no counterpart" omits the other two declarative-only members, onColumnAdd?: never and onCardAdd?: never (complex.ts:165/173, retired tombstones) — the "7" counts them.
9. A3 (ObjectKanbanProps from @objectstack/spec/ui): 13 keys; shares 11 with the plugin's 19; cannot express 8 (allowCollapse cardTemplates className columnWidths limit onCardMove onQuickAdd type); adds 2 (filter titleField). PR body reproduces exactly — RIGHT.
10. Latency: SchemaRegistry\s*\[ repo-wide (excl. node_modules, dist): 0 non-pin hits; 6 hits, all in the PR's two pin files; control ComponentPropsMap\[ 32. Generic SchemaRegistry[K] outside tests: 0. Sibling /home/user/objectstack: 0 imports of SchemaRegistry from @object-ui/types (control: 851 mentions of the bare name there). "Nothing in this repository performed such an indexed access" — RIGHT.
11. The PR's pins are lit and in-program. Types pin via tsc -p packages/types/tsconfig.test.json --listFiles: exit 0, pin 1, registry.ts 1, 407 files. Plugin pin isolated through tsconfig.test.json settings: exit 0, dist 1 / src 0. My three ablation legs, predictions written first, all hit: A restore 'kanban': DeclarativeKanbanSchema (+ its import) → exit 2, exactly 1 error, test.ts(77,7); B delete the key → exit 2, 5 errors, (72,28) + TS2339 at 77/83/89; C add a member the plugin lacks, rebuild dist (marker 0→1→0) → plugin pin exit 2, 1 error (69,7), back to exit 0 after restore-and-rebuild. Every mutation proved by git hash-object movement and anchored counts, every restore by blob equality and empty git diff HEAD -- packages/. RIGHT.
12. Changeset's account of the gap. "objectui#6172's ruling kept the plugin's bare names rather than relocating that dialect down here" — RIGHT (comment 5474954379, 2026-08-31, option A; option C "promote the plugin's dialect INTO types" was on the table and not taken). "…so the gap is permanent by decision" — WRONG on the record. finding(types/plugin-kanban): an authored kanban document can PASS safeValidateSchema and still render empty — the validator and the renderer honour different faces #7664 comment 5548643216 (director seat, maintainer verbatim 「同意」, batch Add 8-directional resize system and premium Tailwind design to designer #41, 2026-09-05T02:09:40Z) rules (a): the @object-ui/types 'kanban' arm is rewritten to the plugin's shape, plugin-kanban imports it, registry.ts is in the execution list, state pm:queue. That is the relocation the changeset calls permanently declined. The phrase was in the changeset from 41ff58ff7 (line 23) and was carried unchanged into 3748de9af (line 29), authored 02:18:01Z — 8m21s after the ruling; it also stands at registry.ts:201 and the types pin docblock line 32.
13. Scope of the remediation, by measurement. git diff-tree -r --name-status 41ff58ff7 3748de9af → exactly one path, M .changeset/7645-schema-registry-kanban-honesty.md. Tree ids for packages, packages/types, packages/plugin-kanban, scripts, apps identical between the two commits; .changeset tree id moved ebcdbec46e66 → dba7d9603c61 (the control that proves the comparison can read a difference); root trees d33b9845… → 63928a33…. The "one markdown file, no source/test/behaviour change" claim is RIGHT. Whole PR vs merge-base: 4 paths; packages/plugin-kanban/src/types.ts, packages/types/src/complex.ts, packages/types/src/base.ts blob-identical; nothing under content/docs/releases/ (from the name-status listing; my tree-id probe of that path was unresolvable at both commits and is discarded as a dark reading).

On the prior REFUSE: every one of its twelve derived judgments and its three changeset defects reproduce under my instrument; none was wrong. The three defects are fixed at head (item 8: three shared names; item 3: the break named with TS2741; item 4: the any-read mechanism named). My measurement does not disagree with it anywhere.

② Semver grading

Declared: '@object-ui/types': minor (changeset line 2). Level: correct. AGENTS.md §版本号策略 (line 239): 「changeset 里不要声明 major … objectui 自身的破坏性变更也标 minor(在正文里写清 breaking 语义即可)」, enforced by scripts/check-changeset-no-major.mjs: exit 0 at head (✅ No changeset declares a \major` bump.); lit control with a planted majorchangeset → exit 1 with the gate's own❌ A changeset declares a `major` bump(control file removed, tree clean). The rule's second half — spell the break out in the body — is now satisfied for both measured effects (the "Breaking, loudly" bullet with TS2741, the "Silent, for member reads" bullet with the index signature, and the closing paragraph naming it a breaking change shipped asminor). @object-ui/plugin-kanbancorrectly has no entry (unpublished test file only; samefixedgroup). CIChangeset Bump Policy, Changeset Declaration, Changeset Overwrite Report`: success at head. The one defect in the body is item 12 — a factual claim about the gap's future, not a bump claim.

③ Boundary-flag disposition

• F1 — extend the one-authority family rule to every SchemaRegistry key? (card's question; PR body's "one question left for the PM"). PM ruled B now, C as follow-up; SchemaRegistry's docblock promises more than the layer can deliver — it owns the key set, not the values #7665 exists (pm:queue, 2026-09-04T16:02:00Z). Answered; not this PR's.
• F2 — which face is authoritative for an authored kanban document? (dev report Q2; filed as finding(types/plugin-kanban): an authored kanban document can PASS safeValidateSchema and still render empty — the validator and the renderer honour different faces #7664.) Explicitly deferred — it is the maintainer's, and the maintainer has answered: ruling (a) on finding(types/plugin-kanban): an authored kanban document can PASS safeValidateSchema and still render empty — the validator and the renderer honour different faces #7664 at 02:09:40Z. My grading is confined to whether the PR survives any answer. The change does not pre-empt it: at head both faces flow into the entry (M3, M15), the key is kept (M9), nothing is retired; under (a) the entry is re-pointed with one line; under a hypothetical (b) the _ValueIsNotTheDeclarativeFace pin fires loudly and is deleted with the re-point; under (c) it stays. All loud, none silent — and registry.ts's "⛔ Do not restore … without moving the renderer's dialect into a layer this package may depend on" names (a)'s route rather than forbidding it. The description does pre-empt it, in one phrase: "the gap is permanent by decision" asserts the answer to F2 in the release-note input while the same PR's own report lists F2 as open for the maintainer, and since 02:09:40Z it is false on the record. Not silent — it is written down — but it is exactly the class the prior REFUSE was issued for: a measured-false factual claim in the changeset.
• F3 — check:sdui-registration-pins declared not measured. Accepted: plugin-kanban/src/index.tsx untouched (the plugin's only diff file is the new test); CI Governed Surface Queue Guard success.
• F4 — prior reviewer's recorded miss (entry → plugin compiles): reproduced (M5 passes both trees; M4 fires only at base). No effect on the verdict.
• F5 — finding(types/plugin-kanban): an authored kanban document can PASS safeValidateSchema and still render empty — the validator and the renderer honour different faces #7664's own sequencing note says PR fix(types): SchemaRegistry's kanban entry stops describing a component it cannot name #7662 is "already merged per the card". False: PR is open, draft, merged: false, mergeable_state: behind. Not this PR's defect, but the (a) executor must know whether registry.ts starts from DeclarativeKanbanSchema or BaseSchema & { type: 'kanban' }; the PM should correct the note on finding(types/plugin-kanban): an authored kanban document can PASS safeValidateSchema and still render empty — the validator and the renderer honour different faces #7664. Either way (a) re-points this entry.
• F6 — [Decision] Should ObjectKanban get a real record-source ladder? — KanbanSchema.data is any[], so it has no data binding to resolve #7651 fence. Closed not_planned at 02:09:54Z (ruled B in the same batch). The PR body's Scope and the plugin pin docblock (test.ts:33) still call it "an open maintainer decision" — stale.
• F7 — migration note points at DeclarativeKanbanSchema, which ruling (a) retires as "the second step of that rename". True at head; a release-line sequencing hazard for the central release-notes author if both land in one line. Flag for the PM; not a defect at head.
• F8 — CI at head: 29 checks success, 3 test shards in_progress at read time, nothing red.
• F9 — item 8 precision note (two never tombstones unnamed): optional tidy, not required.

VERDICT: REFUSE

What must change (all wording; the type line registry.ts:218 stays byte-identical and both pins must still read exit 0 through their declared instruments after the edit):

1. .changeset/7645-schema-registry-kanban-honesty.md line 29 — remove the claim of permanence. State the gap as decided by objectui#6172 (2026-08-31) and now scheduled to close under objectui#7664's ruling (a) (2026-09-05), under which this entry is transitional and will be re-pointed at the declared type. No number, bullet, or the bump needs to move.
2. packages/types/src/registry.ts:201 — same phrase ("so the gap is permanent by decision, not by oversight"), same correction; comment-only.
3. packages/types/src/__tests__/schema-registry-kanban-honesty-7645.test.ts:32 — same phrase, same correction; comment-only.
4. packages/plugin-kanban/src/__tests__/schema-registry-kanban-honesty-7645.test.ts:33 and the PR body's Scope line — objectui#7651 is closed (ruled B, 2026-09-05T02:09:54Z), not "an open maintainer decision".
5. PR body — record that finding(types/plugin-kanban): an authored kanban document can PASS safeValidateSchema and still render empty — the validator and the renderer honour different faces #7664 was ruled (a) at 02:09:40Z and that this PR's entry is the transitional state under it, so the second carrier the remediation claimed to fix is honest on the record as well.

Generated by Claude Code

✅ 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 re-review (round 3) — claude-fable-5-1, adopted verbatim

Isolated review subagent. Tier verified from its transcript: 89/89 assistant turns stamped claude-fable-5-1, 0 fallback notices, against a lit control that reads claude-opus-5 in this session's other transcripts. Adopted verbatim — not edited, trimmed or paraphrased.

Third round. Rounds 1 and 2 both refused; this reviewer re-measured both of their grounds and confirms neither was wrong and both are fixed.

Contract review (round 3) — objectui PR #7662 (card #7645), head 55a7470e9

Reviewer: independent, this session; no GitHub writes; shared checkouts read-only throughout (/home/user/objectui HEAD 77b3cf1f5 unchanged, 0 dirty before and after; /home/user/objectstack 0 dirty). Three detached scratch worktrees (head 55a7470e9, previous head 3748de9af, current origin/main 6eebc54b6 fetched first), installed offline from the pnpm store, @object-ui/types built on head and main, plugin-kanban's seven workspace siblings built on head under os-verify-lock.sh (VERDICT command-exit 0); all three trees removed afterwards, 0 review worktrees remain. Toolchain node 22.22.2, pnpm 10.31.0, tsc 6.0.3.

Record state read live, not inherited. PR #7662: open, draft: true, merged: false, mergeable_state: clean; git merge-tree --write-tree origin/main 55a7470e9 exit 0; 32 check runs at head — 29 success, 3 skipped, 0 failed, 0 in progress. Merge-base with current main 2ce2612df; main has moved but touches none of the PR's four paths; registry.ts has no diff merge-base→main, and complex.ts's only main-side hunk is FloatingChatbotConfig (line 922), not the kanban region. #7664: open, pm:queue, unassigned; ruled (a) in comment 5548643216 at 2026-09-05T02:09:40Z (director seat, maintainer 「同意」, batch #41), execution list names registry.ts; its parenthetical "already merged per the card" is false and was corrected in comment 5548799174 at 02:39:35Z, which holds the card until #7662 lands. #7651: closed not_planned at 02:09:54Z, ruled B (5548644219). #6172: closed completed 2026-09-04T15:01:34Z by PR #7643 (merged); ruling 5474954379 at 2026-08-31T07:07:38Z, option A. #7665: open, pm:queue.

Instrument. One probe file (identical bytes both trees, sha fbb25064…), 22 markers, predictions written into the file before either run, compiled through the plugin's own tsconfig.test.json settings so @object-ui/types resolves to dist (--listFiles: dist/index.d.ts 1, src/ 0, plugin types.ts 1, both trees). Four lit controls fired on both trees (TS2344/TS2322/TS2344/TS2322), four non-vacuity pins passed on both. Head: exit 2, 7 errors; main: exit 2, 10 errors — every error at the predicted marker, no miss. One recorded miss elsewhere: my lit-control column prediction (3,24) was inherited from the author's file; my file reads (3,27) — exit, count, code and line all hit.

① Derived judgments

1. SchemaRegistry['kanban']: DeclarativeKanbanSchema → BaseSchema & { type: 'kanban' } (registry.ts:218; published view dist/registry.d.ts:86-87). RIGHT. A1 re-ablated: check:phantom-deps exit 0 clean with its verdict line, exit 1 with one type-only import injected — registry.ts:10:1 [undeclared-runtime] @object-ui/types imports '@object-ui/plugin-kanban' … (type-only); restored to blob b4b0b62a, git diff HEAD -- packages/ empty. A2 from manifests: @object-ui/types 0 runtime workspace deps, plugin-kanban 7 including types — the edge is a cycle. Both faces satisfy the entry at head (M3, M14 pass).
2. Accept set as an assignment target: widened. { type: 'kanban' } TS2741 on main, accepted at head (M11); { type: 'kanban', garbage: 42 } TS2741 on main, accepted at head (M12); { type: 'grid' } TS2322 on both (L4). Changeset names the mechanism. RIGHT.
3. Entry → DeclarativeKanbanSchema: compiles on main, TS2741 at head (Property 'columns' is missing in type 'BaseSchema & { type: "kanban"; }' but required in type 'DeclarativeKanbanSchema', M2). Changeset: "Breaking, loudly, for consumers who indexed SchemaRegistry['kanban']". RIGHT.
4. Member reads collapse to any through BaseSchema's [key: string]: any (base.ts:440). entry['columns']: DeclarativeKanbanColumn[] on main, any at head (M6b/M6a, each the other's cross-control); draggable likewise (M7); undeclared keys any on both (M8); className stays string | undefined on both (M13). Changeset states mechanism, before/after and the undeclared-keys caveat. RIGHT.
5. Plugin face ↔ entry. KanbanSchema → entry TS2322 on main, compiles at head (M4/M14); entry → KanbanSchema compiles on both (M5). The plugin-side pin proves the tag and nothing narrower; the record claims no more. Leg C (below) shows the pin is not vacuous. RIGHT.
6. keyof SchemaRegistry / published ComponentType: unchanged. 69 keys on head and main, lists identical; control — dropping kanban from the head list gives 2 differing lines; Extract<ComponentType,'kanban'> = 'kanban' on both (M9). RIGHT.
7. Exported symbols: unchanged. PR diff vs merge-base has 0 changed lines containing export under packages/*/src; the only non-comment changes in registry.ts are the BaseSchema import, the dropped DeclarativeKanbanSchema import, and line 218. DeclarativeKanbanSchema still exported (index.ts:294) and still the ComplexSchema arm (complex.ts:1212). RIGHT.
8. New pins are unpublished. Both tsconfig.jsons exclude **/__tests__/**; dist contains 0 test paths. RIGHT.
9. Pins lit and in-program through their declared instruments. Types: tsc -p packages/types/tsconfig.test.json --listFiles exit 0, 0 errors, pin 1, registry.ts 1, 563 files. Plugin: tsc -p packages/plugin-kanban/tsconfig.test.json --listFiles on the full project with siblings built — exit 0, 0 errors, pin 1, dist/index.d.ts 1, src/ 0, 1492 files (the unbuilt-sibling run is a precondition failure: 45× TS2307, 0 errors in the pin). Lit control per instrument: a scratch Equal<1,2> file → exit 2, exactly 1 error, TS2344, both. Ablation legs, predictions first, all hit: A restore 'kanban': DeclarativeKanbanSchema (registry.ts from main 6eebc54b6) → exit 2, exactly 1 error test.ts(77,7) TS2344; B delete the key → exit 2, 5 errors: (72,28) TS2344 _KeyKept, (77,7) TS2344 + (77,34) TS2339, (83,60), (89,54) TS2339; C add a member the plugin lacks, rebuild dist (marker 0→1) → plugin pin exit 2, exactly 1 error (69,7) TS2344; restore + rebuild → marker 0, exit 0. Every mutation proved by blob movement, every restore by blob equality and empty git diff HEAD -- packages/. RIGHT.
10. Latency. SchemaRegistry\s*\[ repo-wide (excl. node_modules/dist): 0 non-pin TypeScript hits; 6 hits in the two pins; 5 prose hits (4 in this PR's changeset, 1 in 3965-…). Controls: ComponentPropsMap\[ 32; gate scripts mentioning SchemaRegistry 0 vs packages/types/src 17; sibling /home/user/objectstack 0 imports of SchemaRegistry from @object-ui/types vs 851 bare mentions. "Nothing in this repository performed such an indexed access" — RIGHT.
11. A3 (ObjectKanbanProps, @objectstack/spec/dist/ui/index.d.ts:6617). 13 keys; shares 11 with the plugin's 19; cannot express 8 — allowCollapse cardTemplates className columnWidths limit onCardMove onQuickAdd type; adds filter, titleField. Controls: self 19, empty 0. PR body reproduces exactly. RIGHT.
12. Changeset member census (AST via TypeScript compiler API, head and main identical — plugin-kanban/src/types.ts and base.ts blob-identical head vs main): DeclarativeKanbanSchema 7, plugin KanbanSchema 19, shared 3 (type, columns, onCardMove), column types 6/6 sharing 5; columns required DeclarativeKanbanColumn[] (complex.ts:134) vs optional KanbanColumn[] (types.ts:140); onCardMove type-identical (M15). Controls: DeclarativeKanbanCard 9, self-intersection 7, disjoint pair 0, nonexistent null. RIGHT.
13. Registered renderer. ComponentRegistry.register('kanban', ObjectKanbanRenderer, …) at plugin-kanban/src/index.tsx:434-436; ObjectKanban.tsx:31 imports KanbanSchema from ./types. RIGHT.
14. Wording-only scope 3748de9af → 55a7470e9. TypeScript scanner, trivia dropped: registry.ts 496/496, types pin 231/231, plugin pin 212/212 — 0 differing tokens each. Lit control: a scratch copy with 'kanban' → 'kanbam' on line 218 reads exactly 1 differing token (docs: prioritize 🔮 Next Phase (N.1–N.5) as current active development focus #469 StringLiteral); trivia-kept control reads 12 differing. Line 218 sha256 7fbfa7bd6fd88f6b both heads; line counts 228/94/84 both heads; every changed registry.ts line is a // line (0 non-comment). Blobs moved for all four files. Phrase census head: "permanent by decision" 0 (prev 3), "open maintainer decision" 0 (prev 1), 7664 5 (prev 0), TRANSITIONAL 2 + lowercase 1. RIGHT — matches the PR body's figures exactly.
15. The record's account of the gap. "objectui#6172's ruling (2026-08-31) kept the plugin's bare names rather than relocating that dialect down here" — RIGHT (5474954379, option A). "objectui#7664's ruling (a) (2026-09-05) rewrites this package's 'kanban' arm to the plugin's shape, has @object-ui/plugin-kanban conform to it, and retires the DeclarativeKanban* trio; registry.ts is on its execution list" — RIGHT, each clause is in 5548643216. "so this entry is scheduled to be re-pointed at the declared type … transitional state under that ruling, not the permanent one" — RIGHT: the PM's own hold comment says "fix(types): SchemaRegistry's kanban entry stops describing a component it cannot name #7662's entry is the transitional state under (a)" and "either way, (a) re-points this entry"; finding(types/plugin-kanban): an authored kanban document can PASS safeValidateSchema and still render empty — the validator and the renderer honour different faces #7664 is pm:queue, held for this PR. Nothing claims an executor, a date, or that (a) is dispatched. registry.ts's "Until then" and the pins' "when it lands … this pin retires with the re-point" keep the tense future. Plugin pin's "[Decision] Should ObjectKanban get a real record-source ladder? — KanbanSchema.data is any[], so it has no data binding to resolve #7651 was ruled B and closed as not_planned (2026-09-05T02:09:54Z)" — RIGHT (closed_at matches to the second).
16. Prior reviews re-measured. Review 1's three changeset defects and review 2's one ("permanent by decision", 8m21s after the ruling — 02:18:01Z − 02:09:40Z verified) all reproduce under my instrument; neither REFUSE was wrong. Review 2's item 13 scope claim reproduces (one path). Nothing to overturn.

② Semver grading

Declared '@object-ui/types': minor (changeset line 2). Correct. AGENTS.md §版本号策略 line 239: 「changeset 里不要声明 major … objectui 自身的破坏性变更也标 minor(在正文里写清 breaking 语义即可)」, enforced by check-changeset-no-major.mjs — exit 0 at head (✅ No changeset declares a \major` bump.), lit control with a planted majorchangeset → exit 1 with the gate's own❌ A changeset declares a `major` bump:(control removed, tree clean). The rule's second half is met: the body spells out both measured effects — the loud TS2741 break for consumers who flowed the entry into the declarative face, and the silentanyon declared-member reads via the index signature — and closes by naming it a breaking change shipped asminorunder the alignment rule.check-changeset-presence exit 0 (3 source file(s) of 2 released package(s) … 1 changeset(s)); @object-ui/plugin-kanbancorrectly has no entry (unpublished test file only, samefixedgroup). CIChangeset Bump Policy, Changeset Declaration, Changeset Overwrite Report, Changeset Fixed Group Check`: success at head.

③ Boundary-flag disposition

• F1 — extend the one-authority family rule to every SchemaRegistry key? PM ruled B now, C as follow-up = SchemaRegistry's docblock promises more than the layer can deliver — it owns the key set, not the values #7665 (open, pm:queue). Answered; not this PR's.
• F2 — which face is authoritative for an authored kanban document? finding(types/plugin-kanban): an authored kanban document can PASS safeValidateSchema and still render empty — the validator and the renderer honour different faces #7664, maintainer ruled (a) at 02:09:40Z. This PR does not pre-empt any part of (a): it touches none of complex.ts, zod/*, index.ts, or plugin-kanban/src/types.ts (blob-identical head vs main); no ratchet keys on the registry's value side (0 gate scripts, 0 non-pin indexed accesses); registry.ts's "⛔ Do not restore … without moving the renderer's dialect into a layer this package may depend on" names (a)'s route rather than forbidding it. What (a)'s executor inherits from this PR, enumerated: the one type line registry.ts:218; the BaseSchema import at :9 (unused after re-point — noUnusedLocals: false, @typescript-eslint/no-unused-vars is warn, so a tidy, not a gate); the comment block :187-217; and both pin files, which import DeclarativeKanbanSchema and therefore retire with the trio — disclosed in their own docblocks. _KeyKept is worth carrying forward. Clean.
• F3 — check:sdui-registration-pins declared not measured. Accepted: index.tsx untouched; CI Governed Surface Queue Guard success.
• F4 — check:phantom-deps not re-run for 55a7470e9 (declared narrowing). Closed: run here, exit 0 with verdict line, plus the A1 ablation.
• F5 — finding(types/plugin-kanban): an authored kanban document can PASS safeValidateSchema and still render empty — the validator and the renderer honour different faces #7664's "already merged per the card". Both statements verified: the ruling's text says it; the PR is open/draft/unmerged. Corrected on the card at 02:39:35Z, 14 min before commit 55a7470e9; the PR body's "for the PM, not edited here" line does not mention the correction — true, incomplete, harmless.
• F6 — [Decision] Should ObjectKanban get a real record-source ladder? — KanbanSchema.data is any[], so it has no data binding to resolve #7651 fence. Fixed at head (plugin pin :33, PR Scope).
• F7 — migration note names DeclarativeKanbanSchema, which (a) retires. True at head; the PM's hold sequences (a) after this PR, so the central release-notes author sees two records on this entry in order. Note, not a defect.
• F8 — precision, optional: (i) the changeset's "no counterpart" sentence omits the two never tombstones onColumnAdd/onCardAdd the "7" counts; (ii) the PR body's "exactly 1 hit" latency count predates the pins and the expanded changeset — at head it is 0 non-pin TypeScript hits, 6 pin hits, 5 prose; substance holds; (iii) the finding(types/plugin-kanban): an authored kanban document can PASS safeValidateSchema and still render empty — the validator and the renderer honour different faces #7664 ruling says "18-member KanbanSchema" while the changeset's measured body count is 19 — the changeset is right for what it scopes; the (a) executor should inherit neither number.
• F9 — CI / mergeability at head: all green or skipped; mergeable_state: clean; merge-tree clean.

VERDICT: PASS

Generated by Claude Code