docs(types): cite the retire-vs-remove discriminator once instead of restating it at eight sites (objectui#9684)

[os-sales]

Part of #9684 — this PR takes the live half and hands the rest back as a decision; the card stays open for it.

RE-DERIVED AT HEAD baa386c4b — one vintage, no stale sentence left standing

Written by the implementer in the same push that created this head. ⭐ Round 5 established the rule this body follows: the whole body is ONE vintage, every figure re-derived against the head being pushed, and a sentence the review falsified is repaired in place rather than annotated. ⛔ Nothing is left for a banner to excuse.

What round 5's review falsified, and what became of it here:

1. "caught only directly behind or, with nothing but a comma, whitespace and an opening parenthesis allowed in between" — that named a set while the pattern ,?\s*\(? required an order. ⛔ Repaired by moving the pattern, not the sentence — see A below. The sentence now reads "any number of them, in any order", and it is true.
2. "Its population is every comment in the package" — two ABUTTING comments were one block, so a drifted copy inherited its neighbour's precondition. ⛔ Repaired in the code — see B below.
3. "Ablation ledger — 17 legs" beside a paragraph enumerating 18. ⛔ Repaired by re-deriving the number from the table, which is now 41 and is counted from the rows below rather than remembered.

⚠️ Round 5's body also carried this sentence: "⛔ The clause pattern itself is unchanged; nothing was widened to make a sentence true." That is no longer true, and round 6 reverses that choice on purpose. The reversal and its four-axis reasoning are stated below rather than quietly dropped.

Clause-②: yes

What round 6 changed

Round 5's contract review passed the direction for the fifth time and reproduced every figure the round printed — the pin, the reverse leg, the full suite, the 65-file .d.ts equality with both marker controls, the population at base and head, and the exact failure sets for B1, C1, C2 and D1. Standing A, standing B/③C and ③D are all closed. Two findings remained, both in the silent direction, both wordings nobody had tried.

A · the separator sentence named a SET; the pattern enforced an ORDER. All four carriers said 3, iii, third and thirdly are caught behind or "with nothing but a comma, whitespace and an opening parenthesis allowed in between" — a set of permitted characters. The pattern spelled that separator ,?\s*\(?, which admits comma, then whitespace, then parenthesis, in that order only. So , or ( 3 ) … (whitespace after the parenthesis) and , or ,3 … (a comma after whitespace) contained nothing any published sentence forbade and passed 22/22.

⛔ The sentence was not patched to describe the order. The separator is now [\s,(]* — the genuine set its prose always claimed — and the caught-list in all four carriers is re-derived from the widened pattern, 20 wordings re-measured through the pin (legs W01–W20), never hand-patched. Two properties make that safe and checkable:

• [\s,(]* is a strict superset of ,?\s*\(? — every string the old separator admits is made only of commas, whitespace and opening parentheses ⇒ ⛔ no wording that was caught became silent. Measured, not argued: the eight previously-caught wordings are all still caught (W09–W15, W20), and the four wordings the header declares silent are all still silent (W05–W08).
• The two escapes now red (W01, W02), and so do the two further set-shaped spellings nobody had published either: , or, (iii) … (W03) and , or (( 3 )) … (W04).

⚠️ Two observations the review asked to be folded in, both measured here: , and prong-3 keeps … is silent (W19) — the hyphenated spelling is not prong 3 and ⛔ no carrier claims it is; and , or third-party tooling keeps … reds (W20) — the anchor catches a non-numbering third behind or. That second one is the declared spurious-red direction and is fine by design: a spurious red sends a human to read four blocks, a spurious green is the defect this pin exists for.

B · two abutting comments were ONE block, so a drifted copy inherited its neighbour's precondition. A blank line was the only thing that ended a comment run. So an un-amended // statement wedged between the ChatbotSchema JSDoc's closing delimiter and export interface — no blank line — passed 22/22, and so did an un-amended JSDoc stacked directly above that JSDoc. A human reads five statements; the pin counted four.

⭐ The pin's own prose already forbade exactly this: "A blank line ends a run on purpose — joining across one would let a drifted note inherit a compliant neighbour's precondition and pass." The code joined across */→// and */→/** anyway. ⛔ The prose was right; the code now says it — the same direction ③D took. A run now ends at a */ and starts at a change of kind.

⚠️ That is a boundary rule, ⛔ not a rule about statements, so it re-cuts every abutting pair in the package. Re-derived at this head with the pin's own comment oracle rather than remembered:

abutting shape, no blank line	sites in packages/types/src
*/ → //	3 (spec-ui-namespace.ts:32, zod/objectql.zod.ts:679, zod/objectql.zod.ts:1904)
*/ → /**	13
*/ → /*	1 (field-types.ts:903)
// → /**	11 (complex.ts ×2, index.ts ×5, layout.ts, objectql.ts ×2, zod/objectql.zod.ts)
// → /*	1 (index.ts:1326)
total	29

The block count over the package goes 4696 → 4725 — one new cut per pair, exactly — and ⛔ no count anything asserts moves: the same 4 statements at the same four line spans (complex.ts 165, 1026, 1381, 1830), 9 trailing comments, 0 of them stating the rule, before and after.

⚠️ A figure in the round-5 handoff that this round could not reproduce, stated rather than quietly corrected. The handoff recorded this census as "*/→// with no blank 0, */→/** with no blank 0, //→/** with no blank 5 (index.ts 1085, 1129, 1170, 1224, 1428)". Measured here package-wide, those are 3, 13 and 11; the named 5 are the index.ts subset of the third row, and every site listed above was read back by eye with cat -A before being reported. ⇒ the fix was verified against the real 29, ⛔ not against the 5.

What the card asked for, and what the tree says now

The card measures one live defect on PR #9682's head dce4a74a1: the sixth statement of the retire-vs-remove discriminator, the GestureConfig note in mobile.ts, did not carry the amended surviving-carrier precondition.

⚠️ That instance is already repaired, and it was repaired before #9682 merged. #9682 landed as 5f8190c8cc98409e65400ff24adfb805b80a3ffc at 2026-09-17T11:56:43Z — thirty minutes after the card was filed — and its merged diff amends the GestureConfig note along with the other four. Read on this branch's base fc12bc8c6, that note states the amended rule in full.

⭐ The card's class is live anyway, at two sites the card did not name. Re-derived on fc12bc8c6, which is still this branch's merge-base with origin/main at this head:

file @ fc12bc8c6	retirement it argues	carried the precondition?	in the card's six?
complex.ts block opening at :165	KanbanColumn.color	yes	yes
complex.ts block opening at :1026	ChatbotSchema dark keys	yes	yes
complex.ts block opening at :1381	displayMode	yes	yes
complex.ts block opening at :1830	triggerIcon	yes	yes
mobile.ts block opening at :74	MobileResponsiveConfig	⛔ no	⛔ no
mobile.ts block opening at :122	MobileOverrides	yes	yes
mobile.ts block opening at :293	GestureConfig	yes (by 5f8190c8c)	yes
mobile.ts block opening at :371	MobileComponentConfig	⛔ no	⛔ no

Every address above is bound to fc12bc8c6; none of them is carried into any file in this diff. The same population read at this head is four, all in complex.ts — which the pin asserts rather than this table.

So the population was eight, not six, and the amendment reached five of eight. ⇒ the card's own argument holds harder than it claimed: the failure is not "one site was missed", it is "the instrument that enumerates the sites is itself a phrase grep, and it undercounts".

The widened scan, and its controls

⛔ The card's six were not treated as a closed list. Scans, all bound to fc12bc8c6:

• Concept scan, whole tree (not just the two files, not just the package): git grep -inE "steers? authors to|named live replacement|keeps? loud|kept loud|taught as working|surviving carrier|no carrier" over every tracked file. Lit control: it fires well outside the two files — packages/app-shell/.../FlowNodeConfigField.entryCondition.test.tsx, packages/types/src/__tests__/page-breadcrumbs-refusal-8871.test.ts, three more test files, scripts/check-governed-queue-guard.mjs, and five .changeset/*.md bodies (4919, 5942, 7519, 7678, 9183) — so its zero elsewhere is a reading, not a silent probe. ⚠️ Rounds 1–5 printed "four .changeset/*.md bodies" here; re-derived at fc12bc8c6 in this push it is five, and the list is now named rather than counted from memory.
• Route-argument scan, for paraphrases that state the rule without any of those phrases: git grep -inE "rather than (a )?tombston|rather than kept as|removed outright|outright rather than|than a plain removal|never. carcass|tombstone (exists|is available|rather)" over packages/types/src excluding tests and zod/. This is the scan that found MobileComponentConfig, which paraphrases the precondition as "There is no surviving object to hang a never key on" and is invisible to every phrase the card's probe searched for. Lit controls in the same pass: it fires on ai.ts and on objectql.ts, neither of which the card names.
• Two sites outside the card's two files, reported and deliberately NOT changed: AIFormAssistSchema.formId in ai.ts cites "PRONG 2 of the discriminator (objectui#5941, feat(types,mobile)!: retire the MobileComponentConfig published type #7526, types: the triggerIcon tombstone's rationale contradicts mobile.ts's own retire-vs-remove discriminator, and its tripwire comment overclaims #7678)" without restating it — it is already the shape this card asks for, and it needed no edit when types: the triggerIcon tombstone's rationale contradicts mobile.ts's own retire-vs-remove discriminator, and its tripwire comment overclaims #7678 amended the rule, which is the measured precedent for the remedy below. ObjectKanbanSchema.allowCollapse in objectql.ts argues "why a tombstone rather than a plain deletion" from the passthrough silent-keep mechanism; that is a different argument, not this discriminator, so it is a boundary case, not a ninth site.
• ⚠️ A wrapping hazard the card's probe shares with any line-based grep: grep -inE reads one line at a time, and this prose wraps mid-phrase. Live demonstration re-taken at this head on a sentence this PR must not disturb: grep -c 'no silent-strip hazard for prong 2 to guard' over mobile.ts prints 1, while flattening the comment punctuation and line wrapping first prints 2 — the second occurrence wraps. The pin flattens before matching for exactly this reason.

The change

mobile.ts now cites the rule instead of restating it. The statement it cites is the ChatbotSchema JSDoc in complex.ts; each of its four retirement notes points there and keeps its own per-retirement argument — which prong holds here and why — which no shared statement can carry, exactly as the card says. ⚠️ That is the statement they cite and ⛔ not the only one in the package: complex.ts states the rule at four sites, ChatbotSchema among three peers (KanbanColumn.color, displayMode, triggerIcon), and whether those four become citations too is the open decision below. Nothing is deleted into a vacuum: every prong measurement, reopen condition, spec-history note and ⚠️ carve-out is kept. The MobileComponentConfig note keeps its own precondition as the precondition rather than dropping it, with its wording brought to the amended form — "no surviving object" became "no surviving carrier", which is the rule's own phrasing and what the pin matches.

A file preamble states, once, where the rule lives and why this file cites while complex.ts states.

Measurements — all re-taken on baa386c4b

① The .d.ts asymmetry the card lists as "not measured". mobile.ts argues its retirements in // comments; complex.ts argues its four in JSDoc on exported members. Declaration emit treats those differently. Measured by injecting one marker into each and rebuilding the package clean, under the verify lock:

marker injected into	occurrences in the emitted declaration file
a // note in mobile.ts (present once in src in the same run)	0 in dist/mobile.d.ts; per-file hash table unchanged vs the head table (0 diff lines)
the ChatbotSchema JSDoc in complex.ts (present once in src in the same run)	1 in dist/complex.d.ts, 0 in dist/mobile.d.ts; per-file table moves 4 diff lines

⚠️ A first attempt at these two legs measured nothing and is reported as such. Both replacements re-emitted their own anchor, so ablation-replace refused the mutation ("the anchor count moved 1 -> 1, a drop of 0") and the build that followed carried no marker at all — a reading of 0 that would have read exactly like the real leg. ⛔ It is not counted as a measurement; the legs above are the re-run with consumed anchors, each with its src count printed beside its dist count so a zero cannot be an unplanted marker. Residual markers after the pass: 0 files in src/, 0 in dist/.

⇒ a citation costs a consumer of the published declarations nothing in mobile.ts, and would cost one a jump out of the block they are in for every site in complex.ts. That is the reason this PR converts one file and not the other.

② What a .d.ts consumer's reading changes to: nothing. Clean full rebuilds in one locked pass (rm -rf dist + pnpm --filter @object-ui/types build, then per-file sha256sum over dist/**/*.d.ts), never incremental: this head emits 65 declaration files, 130 files in dist total; with this branch's base mobile.ts (blob 5e3ad63df) swapped in it emits 65; the per-file hash tables differ by 0 lines and the set-hash is identical (6774b4db26140485). A third clean build after restoring the head file reproduces the first table exactly — 0 diff lines — so the equality is ⛔ not a stale-dist artefact. The instrument is the same one that scores 1 and moves the table by 4 lines for the complex.ts injection above, so it is ⛔ not a probe that cannot fire.

③ Reverse verification of the pin, at this head. Fix committed first, then mobile.ts alone written back to the base blob under a trap ... EXIT INT TERM, the pin re-run, then restored with git checkout HEAD -- ABSOLUTE_PATH (⛔ never a bare git checkout --, which restores from the index):

• this head: 22 passed / 22.
• base mobile.ts swapped in: 8 failed / 14 passed. The precondition assertion and the outside-complex.ts assertion both name exactly the four base blocks — mobile.ts blocks opening on lines 74, 122, 293 and 371 — found by an instrument that was not told where to look; the four citation assertions and both mobile.ts statement assertions fire alongside them. ⭐ Unchanged from round 5, which is the point: the boundary change re-cut 29 pairs and moved this reading by nothing.
• Landing proven by a changed blob (1d33aebaece0 → 5e3ad63dfb29) and by an on-disk count of a base-only phrase, ⛔ never by an exit code. Restore verified by blob hash (1d33aebaece0 back to HEAD), an empty git diff HEAD and an empty git status --porcelain.

The pin

packages/types/src/__tests__/tombstone-discriminator-agreement-9684.test.ts derives the population of rule statements from the source at run time. Six describe blocks, 22 assertions at this head — read off the file (16 it / it.each declarations, two of them .each over 4 cases). It fails when

1. any comment in this package states the rule without its surviving-carrier precondition in the rule's own phrasing,
2. any file in this package outside complex.ts states the rule at all — a prong, a clause, anything that matches its statement markers, in a comment of any form, including one riding on a code line and including one glued directly against a compliant neighbour,
3. any of mobile.ts's four notes stops citing ChatbotSchema, or the citation target stops stating the rule,
4. complex.ts states the rule at any number of sites other than the recorded four, or at a site the roster does not name,
5. one of those four carries a clause its peers do not, and
6. the pin itself quotes the rule it guards.

Its population is every comment in the package. A comment run ends at a blank line, at a */ and at a change of comment kind, so two abutting comments are two blocks; a comment riding on a code line is its own one-line block. It carries its own controls: a synthetic un-amended statement that must be flagged, a synthetic amended one that must not, prose that merely mentions a tombstone that must stay out of the population, and two vacuity controls — the four published statements in complex.ts must still be found, and the roster must still resolve to four distinct blocks, so a marker that stops matching reds instead of quietly emptying every other assertion. ⛔ What it cannot do is named in its header rather than hidden: a NEW paraphrase past every marker is invisible to it, exactly as MobileComponentConfig was to the card's grep. "Is this span a comment" is answered by scripts/js-comment-mask.mjs, never by a private regex.

Ablation ledger — 41 legs, every one restored and proven

⛔ No sentence above about what the pin catches is asserted; each is a hypothesis run through the pin. 41 = 6 pre-fix + 21 separator + 4 adjacency + 5 regression + 1 reverse + 4 .d.ts — counted from the rows below, ⛔ not remembered. Every source mutation went through ablation-replace (literal anchor that must hit a declared number of times, on-disk landing proven by a changed blob, restore proven by blob equality plus an empty git diff HEAD). Working tree after every batch: git status --porcelain 0 lines.

⭐ Six mutations were REFUSED by the tool and measured nothing — their replacements re-emitted their own anchor, so the anchor count did not drop and ablation-replace exited non-zero before any test ran. They are named here rather than silently re-run: the first attempts at E4, E5, B1, C1, C2 and the two .d.ts marker legs. ⛔ None is counted as a leg; each was re-run with an anchor the replacement consumes, and those re-runs are the rows below. That is the tool doing the job it exists for — a replacement that does not land produces a green that reads exactly like a pass.

Pre-fix — the two escapes reproduced at e39b1a8da, the head round 5 reviewed

leg	mutation	mutated blob	pin
E1	, or ( 3 ) … on the ChatbotSchema statement	3dd7dc9432a9	22/22 — silent
E2	, or ,3 … on the same statement	7858d7d5f1b5	22/22 — silent
E3	un-amended // statement between the ChatbotSchema JSDoc's */ and the declaration	2e66e693f106	22/22 — silent
E4	un-amended JSDoc stacked directly above that JSDoc, no blank line	deb6c5d5fbe7	22/22 — silent
E5	control: the same stacked JSDoc with a blank line	7568f7ec28b3	4 failed
E6	un-amended // statement above that JSDoc, no blank line	46da257f395e	2 failed

E1 and E2 reproduce the review's blobs exactly. E5 is what says the instrument was not simply blind: separate the blocks by one blank line and it reds 4.

Separator wordings — all 21 re-measured at this head, the caught-list re-derived from the widened pattern

leg	clause appended	blob	pin	published as
W01	, or ( 3 )	3dd7dc9432a9	1 failed	caught
W02	, or ,3	7858d7d5f1b5	1 failed	caught
W03	, or, (iii)	043b990f235c	1 failed	caught
W04	, or (( 3 ))	930694c9d41a	1 failed	caught
W05	, and (iii)	2a16978c2c0a	22/22	silent
W06	, or else (iii)	87ab694c7eca	22/22	silent
W07	; and, third,	d7e68454cc5c	22/22	silent
W08	, or: (iii)	4f39e633dc00	22/22	silent
W09	, and (3)	7ac7178b7d6d	1 failed	caught
W10	, and prong 3	97e341d72bcf	1 failed	caught
W11	; and the third prong	d3a2e2cf7f24	1 failed	caught
W12	, or thirdly	04e633935e3f	1 failed	caught
W13	, or else (3)	1f6a3808f51f	1 failed	caught
W14	; or 3)	607141edf530	1 failed	caught
W15	, OR (III)	ad27035ae7dc	1 failed	caught
W16	, or—(iii) (em dash)	bb6083c2d4f2	22/22	silent
W17	, and third,	3483d58c0a7e	22/22	silent
W18	, or (numbers nothing)	8933075b669b	22/22	silent
W19	, and prong-3	c62d60c5a29a	22/22	silent
W20	, or third-party tooling	e7c5c7c8a2af	1 failed	caught — declared spurious red
W21	, or / (iii) split across a line break	04b3bc24932c	1 failed	caught, because markers match the FLATTENED block

Every red names one assertion: the ChatbotSchema statement carries exactly its recorded clauses, listing a-prong-beyond-the-two present at one site and absent at its peers. ⭐ W05's blob 2a16978c2c0a, W21's 04b3bc24932c, W14's 607141edf530 and W15's ad27035ae7dc are the same blobs rounds 4 and 5 printed for the same mutations — an independent cross-check that these are the review's legs and not lookalikes.

Adjacency — the same four mutations as E3–E6, re-run against the fixed pin

⭐ The blobs are identical to the pre-fix rows, so nothing about the mutation changed; only the pin did.

leg	mutation	blob	pin before	pin after
X1	*/ → un-amended // statement → declaration	2e66e693f106	22/22 silent	5 failed
X2	un-amended JSDoc stacked above, no blank line	deb6c5d5fbe7	22/22 silent	4 failed
X3	control: the same, with a blank line	7568f7ec28b3	4 failed	4 failed
X4	un-amended // statement above the JSDoc, no blank	46da257f395e	2 failed	4 failed

⭐ X2 and X3 now fail identically — that is the whole fix: a blank line no longer decides whether a drifted copy is seen. X4 was already caught and still is; it now reds through the substantive assertions (exact count, roster names, shared clauses, precondition) instead of only through the two kind === 'jsdoc' assertions, and the 11 real //→/** sites in the package are untouched by it — measured as the unchanged population above (4 statements, 9 trailing, 0 stating the rule) and by the pin's own 22/22 at this head.

Regression — the legs the earlier rounds closed, re-run to show the boundary change moved none of them

leg	plant	blob	pin
B1	fully amended statement in data-display.ts	f4e7332729c7	1 failed, and exactly one: no file outside complex.ts states the rule AT ALL
B2	same, precondition removed	1b078ce7682d	2 failed: that assertion plus the precondition assertion
C1	round 4's exact wording appended to the GestureContext line in mobile.ts	2ce74bc7ceb5	4 failed
C2	amended statement appended to the ChatbotSchema declaration line	396d674178ed	2 failed: both complex.ts roster assertions
D1	vacuity control: every statement marker replaced with one that matches nothing	a8152fa73877	9 failed / 13 passed, including both vacuity controls and all four roster assertions

⇒ a dead marker cannot print the same quiet zero B1's assertion prints when the tree really is clean. All five readings are identical to round 5's.

Reverse and .d.ts — 5 more

The reverse leg (③ above) and the four .d.ts legs: base-mobile.ts swap, third-build reproduction, the ChatbotSchema JSDoc marker control, the mobile.ts // marker control.

Why this shape and not the others — the four axes

Option A — amend the two stragglers in place (the shape the card's remedy section refuses). Option B — one block, every site cites it. Option C — cite where a declarations reader pays nothing, state where it pays, pin the agreement (this PR). Option D — pin only, leave all eight statements.

实际业务需求. Measured, not assumed: the rule is stated in six of this repo's changeset bodies — 4919, 5942, 7519, 7678, 7703 and this card's own 9684, the list that marker sweep prints rather than a remembered count, re-derived at this head — and it was amended once (#7678), an amendment that reached five of eight sites. So the recurrence is real. But the cost actually paid was a reader getting the WRONG rule, not an author paying for a slow edit ⇒ agreement is the demand, fewer copies is a means. And the consumer side is asymmetric by measurement ①: a .d.ts reader has never been able to read mobile.ts's four. A remedy that charges that reader a jump to fix a problem they do not have fails this axis, which is what separates C from B.

项目长远合理性. N statements of one rule is the N-dialects shape AGENTS.md #0.1 refuses for metadata, and prose inherits the reasoning: one contract beats N. A is the patch that fails it outright — it grows N to eight-agreeing and makes the ninth amendment a nine-site edit. B is right in principle and wrong on the published face, because complex.ts's statements ARE part of what the package publishes. C takes the principle exactly as far as the contract face allows and says so in the file. D leaves the structure that produced this card intact.

防 AI 写代码犯错. This card exists because an agent applied an amendment at five of eight sites and nothing went red — the declaration was not enforced. The pin is the 声明即强制 half, and round 6 is that half applied to itself for the third time: the pin's prose promised a reach its code did not have, and the code was moved to the prose rather than the prose trimmed to the code (scope B, population ③D, and now the block boundary). ⚠️ Its remaining reach is named rather than implied: enforcement is real for every clause the roster records and for every statement any marker matches; a clause worded past all of the markers is silent, and that silence is stated as the pattern rather than summarised. ⛔ No consumer-side tolerance anywhere: nothing here accepts a looser spelling, and the pin refuses by name.

创业阶段不扩散需求 — ⚠️ and this is where round 6 reverses round 5. Round 5's body argued: "⛔ the clause pattern was not widened to make a sentence true, because widening it would have owed a re-derived and re-published caught-list, which is capability growth bought to retire a prose defect." ⛔ That reasoning was wrong, and it is withdrawn here. The 创业阶段 axis governs declared capability surfaces with no pull — an authorable key, an exported symbol, a published spelling. a-prong-beyond-the-two is none of those: it is an internal test guard, it publishes nothing (measurement ②), and widening it is a contract tightening, which the 防 AI 写代码犯错 axis prefers outright (契约收紧优于消费端宽容). ⇒ the earlier reading treated "more characters in a regex" as capability growth, and that is a category error. The cost round 5 was unwilling to pay — a re-derived and re-published caught-list — is paid in full in this push, by measurement, 20 wordings through the pin. ⭐ And the pin's own header already carried the instruction: "when you widen a pattern, RE-DERIVE this paragraph from the pattern rather than editing the sentence that summarises it." Round 6 follows it instead of quoting it. The rest of the axis holds unchanged: no staged window, per the 2026-08-27 ruling — mobile.ts converts in this one change, with no second spelling left behind and no grace period.

⇒ C. A and D are refused on axis 2 and 3 respectively; B is refused only for complex.ts, and only on evidence that did not exist before measurement ① was taken.

⚠️ Decision box — the half this PR does NOT take

complex.ts still states the rule at four sites, all published. Converting them is the trade the card declines to decide, and measurement ① is what makes it decidable:

• Convert them too (cite ChatbotSchema from the other three): the next amendment becomes a one-site edit, and every .d.ts consumer reading KanbanColumn.color, displayMode or triggerIcon must jump to ChatbotSchema to learn why the key is a tombstone. Cost falls on a consumer who is not the beneficiary.
• Leave them (this PR): four sites to amend instead of eight, and the pin holds those four to a recorded clause roster, so an amendment landing at one of them reds at the other three — for every clause the roster names, measured by legs W01–W04 and W09–W15. ⚠️ A clause worded past all of the roster's markers is still silent, measured by legs W05–W08 and W16–W19; that is the reach, and it is stated in the test header as the pattern rather than as a summary of it. Recommended fallback if nobody wants to spend a declarations reader's attention.
• ⚠️ Either way, one thing is unresolved and is NOT fixed here because fixing it edits published JSDoc: the canonical block does not say it is canonical. ChatbotSchema's JSDoc attributes the quoted phrase to mobile.ts, which is still accurate, but a reader landing there is not told that this is the one statement the others cite. Naming it belongs with whichever option is chosen (objectui#9930 holds it).

check:changeset-claims asked, so: both pending claims survive this diff

The report-only gate names two pending changesets that describe mobile.ts. 7580-responsive-vocabulary-rehome.md is about where BreakpointName is declared — untouched. 7703-chatbot-dark-keys-retired.md quotes "no silent-strip hazard for prong 2 to guard" and attributes it to mobile.ts; that sentence is still in mobile.ts twice at this head — the count a flattening reader prints, ⛔ not the 1 a line-based grep -c prints, one occurrence being wrapped — and this round's mobile.ts diff does not touch it. Neither claim is born false.

Verification — every figure below taken on baa386c4b

Heavy runs serialized through scripts/pm/os-verify-lock.sh in the sibling checkout (slot objectui-9684-r6); exit codes were captured before any pipe, each into its own variable rather than read off a chained command.

run	result
pnpm --filter @object-ui/types build (clean, seven times — five measuring legs plus two during the refused marker attempts)	exit 0 · 130 files in dist · 65 .d.ts, set-hash 6774b4db26140485, reproducible across builds
pnpm exec vitest run packages/types/ (locked)	exit 0 · 208 files / 4838 tests passed
the pin alone	exit 0 · 22 passed / 22
pnpm --filter @object-ui/types type-check	exit 0
pnpm --filter @object-ui/types lint	exit 0 · 0 errors, 290 pre-existing warnings
node scripts/check-changeset-presence.mjs	exit 0 · empty frontmatter accepted as "releases nothing"
check-changeset-no-major.mjs / -fixed.mjs / -claims.mjs / check-pending-changeset-literals.mjs	exit 0
node scripts/check-control-bytes.mjs	exit 0 · 8000 tracked text files scanned, 85 binary skipped
node scripts/check-new-cross-file-line-citations.mjs	exit 0 · 0 new citations at baa386c4b, enforcement report-only
node scripts/check-test-path-roots.mjs	exit 0 · 2159 filesystem calls in 458 of 3290 test files
node scripts/check-hand-rolled-comment-mask.mjs	exit 0 · no carrier outside DEBT
node scripts/check-comment-mask-corpus.mjs	exit 0 · 5129 files, 1 disagree, within the residue objectui#7882 holds open
node scripts/check-type-check-coverage.mjs	exit 0 · 43/43 packages compile their tests; the pin is inside the program
node scripts/check-governed-queue-guard.mjs --test on the three paths	NOT GOVERNED — 3 paths checked against 5 governed surfaces, none matched

⚠️ Lint scope: this repo's pnpm lint is turbo run lint, per package. All three changed files live in packages/types (plus one .changeset/*.md, which no lint rule reads), and that package's own lint script covers them, so nothing about this diff is outside what ran — no narrowing was needed and none is claimed.

⛔ NOT MEASURED, stated rather than implied: CI's final state on this head — the report is filed when the local runs finish, ⛔ not after waiting for CI to converge. Round 5's head was fully green (33 success / 3 skipped / 0 red, 36 of 36 returned); ⛔ that is a reading about e39b1a8da and says nothing about this one.

Acceptance notes

Out-of-scope observations, noted and not filed — none is a reproducible defect, a broken declared contract, or a trap that makes a runtime refuse authored metadata:

• noted, not filed: the round-5 handoff's abutting-comment census (*/→// 0, */→/** 0, //→/** 5) does not reproduce package-wide; measured here it is 3, 13 and 11, the named 5 being the index.ts subset of the third row. It is a figure in a handoff, not a defect in this tree, and the fix was verified against the re-derived 29. Carrier: this body and the round-6 report, which both print the instrument.
• noted, not filed: the card's measured position table is stale in a way that cost this run a premise check — it was taken on an open PR's head and that PR moved before merging. Carrier: whoever writes the next finding from an in-seat review; no PR or person is currently on this file.
• noted, not filed: AIFormAssistSchema.formId in ai.ts already uses the citation form this card proposes and survived types: the triggerIcon tombstone's rationale contradicts mobile.ts's own retire-vs-remove discriminator, and its tripwire comment overclaims #7678 untouched. It is evidence, not a defect. Carrier: the decision box above, if option B is taken.
• noted, not filed: ObjectKanbanSchema.allowCollapse in objectql.ts argues the tombstone route from the passthrough mechanism rather than from this discriminator. Both arguments are sound; nothing disagrees. Carrier: none.
• noted, not filed: the round-4 review's suggested one-sentence fix for A was itself one step wide (, and (3) … is caught, because (3) carries no or anchor). Round 5's seat has since corrected its own handoff on this point. Carrier: this body and the round-5 record, which both say so.

Labels: ⛔ this round wrote none. needs:contract-review is the seat's to hang or clear; the seat re-hangs it when this head moves, and this push neither hangs, removes nor waits on it.

⚠️ A standing rules conflict, flagged and not silently resolved. This round's dispatch names the PR body re-stamp as an owed item, while the standing os-dev clause says the implementer writes the body once, at PR creation, and never PATCHes it. The conflict is with the maintainer as objectui#9976 and is ⛔ unresolved. This round follows the dispatch, exactly as round 5 did, and says so here rather than choosing a side quietly.

⚠️ Provenance, restated as prose rather than left to a footer block. A PATCH to this body downgrades a session-URL footer to the bare form and unconditionally appends a second footer, so the durable carrier is this sentence: this round was implemented by session_01Xm4WFhEe5mwcgyqHjxR2hn, whose session URL is https://claude.ai/code/session_01Xm4WFhEe5mwcgyqHjxR2hn. The body sent in this edit carries no trailing footer, so the platform's append leaves exactly one.

---

Generated by Claude Code

[github-actions]

changeset-claim-re-read

⚠️ 2 pending changeset(s) describe a file this change touches

Their bodies publish verbatim into the CHANGELOG at the next release, so this is a request to re-read them against your diff — addressed here because you are the one seat that can answer it without re-deriving anything.

⛔ Nothing here blocks, and nothing here is a verdict on your change. This gate exits 0, is not a required context, and judges name resolution, never meaning: it asked whether a pending body names a file you touched. "Is this sentence still true?" is the one question it will not answer, and the one you are being asked to answer.

.changeset/7580-responsive-vocabulary-rehome.md

• names mobile.ts → packages/types/src/mobile.ts — edited by this change

  • BreakpointName (xs…2xl) is now declared in @object-ui/types (mobile.ts) instead of re-exported from the spec. No consumer change: same name, same six members, same export sites on @object-ui/types and @object-ui/mobile. Only its provenance moved. - BreakpointColumnMap is now declared in @object-ui/layout (ResponsiveGrid.tsx), verbatim from the retired $strict schema: six optional column counts, no index signature. responsive-grid's columns input and its resolver are unchanged.

.changeset/7703-chatbot-dark-keys-retired.md

• names mobile.ts → packages/types/src/mobile.ts — edited by this change

  All six have a Zod arm, and that is what decides the route here. BaseSchema is .passthrough() on the Zod side and carries a [key: string]: any index signature on the TS side, so an UNDECLARED key is not refused — it is KEPT. Deleting the members would hand the authored spelling exactly the silent no-op this card exists to close, on both faces at once. This package's retire-vs-remove discriminator (objectui#5941 / feat(types,mobile)!: retire the MobileComponentConfig published type #7526, amended by objectui#7678) leaves that structural hazard to the carrier: where there is no mirror there is "no silent-strip hazard for prong 2 to guard" (mobile.ts). Here there is a mirror to host the refusal, and prong 1 holds by the letter for four of the six. The "deleted" row is pinned live as a control in packages/types/src/__tests__/chatbot-dark-keys-retired-7703.test.ts, so the contrast cannot rot into prose.

Read the paragraph, not the line: both false halves of the objectui#8617 claim sat in one paragraph, and correcting either alone would have left it asserting the same wrong thing.

If a claim did go false, correct the body. That is precedented and prose-only, frontmatter untouched; check-changeset-overwrite.mjs will report the correction as its own case 2 ("correcting a declaration on purpose … legitimate"), which is the intended shape — one gate asks for the read, the other records the write.

Not covered, stated so nobody reads this as more: a born-false claim that spells no line address at all (objectui#9495 coordinated one by ORDINAL — "a grep finds that member first" — and deciding that means reading what the sentence means), a claim spelled as a symbol or a package rather than a backticked file name, and a file named ambiguously.

Compared the checked-out tree with 509f8ed75 (merge-base with origin/main): 2 file(s) changed outside .changeset/, read against 1198 pending declaration(s) that publish a body (1752 pending in total). · run

[github-actions]

✅ Console Performance Budget

Metric	Value	Budget
Eager closure (gzip, 329 chunks)	3050.7 KB	3104.5 KB
Main entry chunk (gzip)	145.7 KB	350 KB
Entry file	index-B3XXVuaJ.js	—
Status	PASS	—

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

Package	Size	Gzipped
app-shell (consoleActionDispatch.js)	0.20KB	0.19KB
app-shell (index.js)	16.69KB	6.21KB
app-shell (runtime-config.js)	20.68KB	7.36KB
app-shell (types.js)	0.01KB	0.04KB
app-shell (urlParams.js)	10.06KB	3.86KB
auth (ActiveOrganizationStorage.js)	25.05KB	9.16KB
auth (AuthContext.js)	0.31KB	0.24KB
auth (AuthGuard.js)	2.07KB	1.00KB
auth (AuthProvider.js)	40.18KB	10.59KB
auth (AuthShell.js)	3.49KB	1.40KB
auth (ForgotPasswordForm.js)	12.21KB	3.45KB
auth (LoginForm.js)	18.15KB	5.39KB
auth (PreviewBanner.js)	0.90KB	0.50KB
auth (RegisterForm.js)	6.65KB	2.22KB
auth (SocialSignInButtons.js)	9.61KB	3.89KB
auth (UserMenu.js)	3.41KB	1.23KB
auth (auth-gate-events.js)	1.29KB	0.66KB
auth (authStyles.js)	5.04KB	1.72KB
auth (createAuthClient.js)	40.21KB	10.80KB
auth (createAuthenticatedFetch.js)	8.46KB	3.43KB
auth (index.js)	3.19KB	1.44KB
auth (invitation-status.js)	1.22KB	0.70KB
auth (org-roles.js)	6.66KB	2.78KB
auth (phone-identifier.js)	1.11KB	0.66KB
auth (types.js)	0.59KB	0.35KB
auth (useAuth.js)	5.30KB	1.02KB
auth (useWorkspaceAdminStatus.js)	11.08KB	4.58KB
collaboration (CommentThread.js)	26.08KB	7.56KB
collaboration (LiveCursors.js)	3.17KB	1.27KB
collaboration (PresenceAvatars.js)	6.49KB	2.64KB
collaboration (PresenceProvider.js)	2.79KB	1.13KB
collaboration (index.js)	1.68KB	0.73KB
collaboration (useCollaborationTranslation.js)	6.05KB	2.52KB
collaboration (useCommentSearch.js)	1.98KB	0.88KB
collaboration (useConflictResolution.js)	7.75KB	1.86KB
collaboration (useMentionNotifications.js)	1.81KB	0.68KB
collaboration (usePresence.js)	6.33KB	1.84KB
collaboration (useRealtimeSubscription.js)	7.91KB	2.01KB
components (index.js)	545.76KB	130.74KB
core (index.js)	8.94KB	3.59KB
create-plugin (index.js)	27.94KB	9.51KB
data-objectstack (index.js)	216.90KB	60.15KB
fields (index.js)	249.22KB	62.85KB
i18n (LocalizationContext.js)	1.76KB	0.96KB
i18n (builtinAggregateLabels.js)	0.86KB	0.49KB
i18n (currency.js)	1.22KB	0.64KB
i18n (fallbackInterpolation.js)	6.25KB	2.77KB
i18n (i18n.js)	8.87KB	3.64KB
i18n (index.js)	5.22KB	2.26KB
i18n (pickLocalized.js)	9.86KB	3.95KB
i18n (provider.js)	32.15KB	10.49KB
i18n (useDisplayLocale.js)	2.85KB	1.45KB
i18n (useObjectLabel.js)	34.34KB	9.17KB
i18n (useSafeTranslation.js)	5.60KB	2.33KB
layout (index.js)	38.83KB	10.95KB
mobile (MobileProvider.js)	0.92KB	0.49KB
mobile (ResponsiveContainer.js)	0.94KB	0.38KB
mobile (breakpoints.js)	1.51KB	0.70KB
mobile (createOfflineDataSource.js)	5.61KB	1.75KB
mobile (index.js)	1.99KB	0.87KB
mobile (offlineQueue.js)	3.91KB	1.35KB
mobile (pwa.js)	0.97KB	0.49KB
mobile (serviceWorker.js)	1.48KB	0.62KB
mobile (serviceWorkerSource.js)	3.41KB	1.48KB
mobile (useBreakpoint.js)	1.54KB	0.65KB
mobile (useGesture.js)	6.96KB	1.98KB
mobile (useOfflineSync.js)	1.99KB	0.72KB
mobile (usePullToRefresh.js)	2.53KB	0.85KB
mobile (useResponsive.js)	0.72KB	0.42KB
mobile (useSpecGesture.js)	5.52KB	2.10KB
mobile (useTouchTarget.js)	1.01KB	0.54KB
permissions (MePermissionsProvider.js)	13.52KB	4.88KB
permissions (PermissionContext.js)	0.31KB	0.25KB
permissions (PermissionGuard.js)	0.89KB	0.45KB
permissions (PermissionProvider.js)	6.24KB	2.16KB
permissions (discardProofCache.js)	1.04KB	0.55KB
permissions (evaluator.js)	8.39KB	3.10KB
permissions (index.js)	0.93KB	0.41KB
permissions (store.js)	0.91KB	0.42KB
permissions (useFieldPermissions.js)	1.28KB	0.53KB
permissions (usePermissions.js)	4.83KB	2.27KB
plugin-ai (index.js)	14.81KB	3.63KB
plugin-calendar (index.js)	49.92KB	14.22KB
plugin-charts (index.js)	71.73KB	20.08KB
plugin-chatbot (index.js)	195.35KB	46.52KB
plugin-dashboard (index.js)	132.96KB	35.17KB
plugin-designer (index.js)	215.94KB	44.33KB
plugin-detail (index.js)	253.63KB	66.00KB
plugin-editor (index.js)	2.23KB	1.05KB
plugin-form (index.js)	136.82KB	34.17KB
plugin-gantt (index.js)	167.62KB	41.26KB
plugin-grid (index.js)	213.44KB	58.21KB
plugin-kanban (index.js)	48.10KB	14.94KB
plugin-list (index.js)	112.74KB	27.70KB
plugin-map (index.js)	21.48KB	6.99KB
plugin-markdown (index.js)	13.88KB	4.80KB
plugin-report (index.js)	43.41KB	11.93KB
plugin-timeline (index.js)	30.07KB	8.74KB
plugin-tree (index.js)	10.58KB	3.72KB
plugin-view (index.js)	85.05KB	21.01KB
providers (DataSourceProvider.js)	0.75KB	0.39KB
providers (MetadataProvider.js)	1.37KB	0.59KB
providers (ThemeProvider.js)	1.90KB	0.85KB
providers (UploadProvider.js)	11.66KB	3.50KB
providers (index.js)	0.45KB	0.23KB
providers (types.js)	0.01KB	0.04KB
react-runtime (index.js)	5.62KB	2.34KB
react (LazyPluginLoader.js)	4.47KB	1.63KB
react (SchemaRenderer.js)	104.82KB	34.67KB
react (data-invalidation.js)	5.05KB	2.08KB
react (index.js)	4.63KB	2.18KB
react (schema-input.js)	4.25KB	2.04KB
react (spec-input.js)	0.20KB	0.18KB
sdui-parser (codegen.js)	6.58KB	2.74KB
sdui-parser (dashboard-widget-options.js)	3.08KB	1.30KB
sdui-parser (index.js)	5.66KB	2.50KB
sdui-parser (input-type.js)	2.84KB	1.40KB
sdui-parser (kanban-quick-add.js)	3.89KB	1.87KB
sdui-parser (parse.js)	25.28KB	7.80KB
sdui-parser (provenance.js)	3.66KB	1.82KB
sdui-parser (types.js)	0.28KB	0.23KB
sdui-parser (validate.js)	14.82KB	4.99KB
types (ai.js)	4.11KB	2.06KB
types (api-types.js)	0.20KB	0.18KB
types (app.js)	2.87KB	1.00KB
types (base.js)	0.20KB	0.18KB
types (blocks.js)	0.20KB	0.18KB
types (complex.js)	2.93KB	1.49KB
types (crud.js)	0.20KB	0.18KB
types (dashboard-filter-alias.js)	6.23KB	2.74KB
types (data-display.js)	3.75KB	1.85KB
types (data-protocol.js)	0.20KB	0.19KB
types (data.js)	0.20KB	0.18KB
types (designer.js)	1.85KB	0.85KB
types (disclosure.js)	0.20KB	0.18KB
types (error-code.js)	1.54KB	0.88KB
types (expression.js)	0.20KB	0.18KB
types (feedback.js)	0.20KB	0.18KB
types (field-types.js)	0.20KB	0.18KB
types (form.js)	0.20KB	0.18KB
types (http-inflight.js)	8.87KB	3.73KB
types (http-retry.js)	4.32KB	2.02KB
types (icon-key-migration.js)	4.26KB	1.63KB
types (index.js)	4.74KB	2.25KB
types (layout.js)	0.20KB	0.18KB
types (managed-by.js)	0.19KB	0.18KB
types (mobile.js)	5.00KB	2.39KB
types (navigation.js)	0.20KB	0.18KB
types (objectql.js)	0.20KB	0.18KB
types (overlay.js)	0.20KB	0.18KB
types (permissions.js)	0.20KB	0.18KB
types (plugin-scope.js)	0.20KB	0.18KB
types (record-components.js)	0.20KB	0.19KB
types (record-semantics.js)	1.28KB	0.67KB
types (registry.js)	0.20KB	0.18KB
types (reports.js)	0.20KB	0.18KB
types (select-option.js)	0.20KB	0.19KB
types (spec-report.js)	5.05KB	1.93KB
types (spec-ui-namespace.js)	0.20KB	0.19KB
types (strict-authoring-face.js)	14.04KB	5.36KB
types (system-fields.js)	3.33KB	1.54KB
types (theme.js)	6.28KB	2.87KB
types (ui-action.js)	8.11KB	3.32KB
types (views.js)	0.20KB	0.18KB
types (widget.js)	0.20KB	0.18KB

Size Limits

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

[os-sales]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: ead58317a404d4871b94d052ea23e982e0c76938

Read at 2026-09-18T17:35:20Z and re-read at 17:50:27Z: head unchanged, draft, 3 files +401/−31, 3 commits, 0 reviews, labels package: types / tests / needs:contract-review. Merge-base with origin/main and with the API base 0fe29cc23 is fc12bc8c6 (17:36:54Z); no commit in fc12bc8c6..0fe29cc23 touches packages/types/src. Every measurement below was taken in a detached worktree of the head under my scratchpad, with pnpm install --offline --frozen-lockfile --ignore-scripts (exit 0, 17:42:20Z); the shared checkout was not edited.

① Derived judgments

Enumerated from the diff (git diff --stat fc12bc8c6 ead58317a, 17:42:34Z: exactly .changeset/9684-discriminator-cited-once.md, packages/types/src/__tests__/tombstone-discriminator-agreement-9684.test.ts, packages/types/src/mobile.ts).

1. mobile.ts: comment-only, no accepted set moved — RIGHT. git diff -U0 on the file shows zero changed lines that do not begin with // (17:42:34Z). stripComments from scripts/js-comment-mask.mjs applied to base and head gives 101 non-blank code lines each, identical; control: base mobile.ts vs base complex.ts on the same instrument reads not identical (17:42:34Z).
2. Published .d.ts surface: no delta — RIGHT, and the dev's byte-identity claim holds. Full tsc -p packages/types/tsconfig.json builds (fresh outDir and fresh tsBuildInfoFile per run, never incremental): head emits 65 .d.ts, set-hash b578c03369acc604; with base mobile.ts swapped in, 65 files, same hash, per-file sha256 diff empty (17:45:54Z–17:46:05Z). Controls in the same pass: a marker in a // note in mobile.ts reaches dist 0 times; the same marker in the ChatbotSchema JSDoc reaches dist/complex.d.ts once and moves the set-hash to 1ee171d124ce5d14 (17:46:15Z–17:46:25Z); a // comment placed directly above an exported declaration with no blank line also reaches dist 0 times (17:46:36Z), so the asymmetry is JSDoc-vs-line-comment, not module-scope-vs-attached. The package's full chain tsc && vite build && check-dist-completeness at head: exit 0, 130 files, 65 .d.ts, same set-hash (17:49:57Z); vite.config.ts emits only dist/zod/index.zod.js, so tsc is the sole declaration emitter and the measurement covers the whole published declaration set.
3. New test file: not a published surface — RIGHT. Under __tests__/, excluded by tsconfig.json; compiled by tsconfig.test.json (tsc --noEmit exit 0, 17:47:00Z); check-type-check-coverage exit 0.
4. New exported symbols / new keys on published payloads: none — RIGHT. The claim-time Clause-②: yes was the conservative direction the ruling requires when the remedy shape is open; measured, the shipped half hangs nothing on a published symbol, and this record overturns the declaration downward: no contract delta.
5. Population re-derived with my own instrument (population.mjs: comment spans from the shared scanner, comment-only line runs as blocks, a concept net wider than the pin's markers; known-positive fires, known-negative silent, fires tree-wide in cli, fields, plugin-detail, plugin-kanban). Generic statements of the rule in packages/types/src including zod/: base fc12bc8c6: 8 (four JSDoc in complex.ts: KanbanColumn.color, ChatbotSchema, displayMode, triggerIcon; four // notes in mobile.ts: MobileResponsiveConfig, MobileOverrides, GestureConfig, MobileComponentConfig), head: 4, all in complex.ts, all JSDoc, all carrying the precondition; mobile.ts at head: 0 statements, 4 citing notes plus the preamble (17:44:01Z). Pre-merge 5f8190c8c^: 7 (the triggerIcon statement was added by docs(types): bring every statement of the tombstone discriminator to the amended form #9682) (17:44:00Z). ai.ts formId cites "PRONG 2 of the discriminator" and objectql.ts allowCollapse argues from passthrough — neither states the rule (read 17:44Z). ⇒ the dev's eight is confirmed; the card's six is not.
6. The card's own probe does not return six. grep -inE "surviving carrier|either prong holds|prong 1|prong 2" on base fc12bc8c6: 4 clusters in complex.ts, 3 in mobile.ts — the third is the MobileResponsiveConfig note via its Prong 1: / Prong 2: lines, absent from the card's table (17:40:03Z); the same cluster is present on 5f8190c8c^, the main immediately before docs(types): bring every statement of the tombstone discriminator to the amended form #9682 merged (17:41:07Z). Both un-amended notes landed on main in 6cc48c4e4 on 2026-09-09T06:26:20Z, eight days before the card. Tree-wide the probe fires in 9 further files (17:40:03Z), so the card's "zero hits outside them" is not a reading either. The card's measurement on dce4a74a1 itself is NOT MEASURED here: that object is not in the local store (17:36:2xZ).
7. The card's named defective site is not defective on this PR's base — RIGHT to say so. 5f8190c8c (merged docs(types): bring every statement of the tombstone discriminator to the amended form #9682, 2026-09-17T11:56:43Z) is an ancestor of fc12bc8c6 and its diff amends the GestureConfig note (17:36:55Z).

② Semver level

• Level: empty frontmatter (releases nothing) — matches the diff. check-changeset-presence accepts it explicitly (exit 0, 17:42:44Z); check-changeset-no-major / check-changeset-fixed exit 0. The zero .d.ts delta in ①.2 is what an empty frontmatter asserts.
• Changeset prose vs diff: "eight sites" — holds (①.5); "all 65 emitted declaration files … identical" — holds (①.2); "fails when one of them omits the surviving-carrier precondition" — holds and is correctly scoped (③.A). ⛔ Falsified: "the one statement lives on ChatbotSchema in complex.ts" — complex.ts states the rule at four sites at head, which the PR's own vacuity control asserts (≥4) and my instrument confirms. Overstated: "keeping its own per-retirement argument … unchanged" and "kept verbatim" — the Prong 1: / Prong 2: measurement lines are verbatim, but the MobileOverrides and MobileComponentConfig arguments were reworded (substance preserved); not load-bearing on its own.
• Pending changesets that name mobile.ts: 1,459 quoted spans across every pending changeset checked against base and head mobile.ts; 0 spans present at base and absent at head; control: the finding(types/plugin-chatbot): six ChatbotSchema keys are declared and read by no registration — loading, showAvatars, userAvatar, assistantAvatar, markdown, height #7703 phrase "no silent-strip hazard for prong 2 to guard" present in both (17:49:24Z). Nothing pending is born false by this diff.
• PR body vs diff: the table, the widened-scan controls (re-run tree-wide at head, 17:48:43Z: lit in changesets, app-shell, plugin-detail, plugin-form, react, scripts), the reverse-verification numbers (6 failed / 6 passed naming the two blocks) and the build/gate table all reproduce. ⛔ Falsified: the decision box's "four sites to amend instead of eight, and the pin makes a miss loud rather than silent" — see ③.A, ablation C.
• Gates run in the worktree (17:42:44Z–17:48:47Z), all exit 0: hand-rolled-comment-mask, test-path-roots, new-cross-file-line-citations (0 new), control-bytes, installed-spec-pin-claims, governed-queue-guard --test (NOT GOVERNED), eslint on both changed source files.

③ Boundary flags

A. What the pin can and cannot fail on (measured, every restore verified by blob hash and empty git diff HEAD, 17:45:22Z–17:45:42Z). Head: 12/12. Neighbour pins mobile-residue-retired-7519 6/6 and chatbot-dark-keys-retired-7703 40/40. Reverse (mobile.ts = base blob 5e3ad63df): 6 failed / 6 passed; the precondition assertion names exactly mobile.ts (block opening on line 74) and (block opening on line 371) — the dev's reading reproduces. Ablation A (un-amended two-prong JSDoc injected on an exported member of complex.ts): 1 failed, caught. Ablation B (precondition deleted from the ChatbotSchema JSDoc only): 2 failed, caught. Ablation D (one citation re-pointed away from ChatbotSchema): 1 failed, caught. Ablation E (a prong re-inlined in mobile.ts): 1 failed, caught. ⛔ Ablation C (a third prong added to the ChatbotSchema statement and nowhere else): 12/12 PASS. The pin guards one clause — the #7678 precondition — not agreement between the four remaining statements. The test header and the changeset say so; the PR body's decision box and the "声明即强制" sentence do not. The next amendment of any other clause is a four-site hand edit whose miss is silent again.

B. open_questions[0] — convert the four complex.ts statements or leave them. Not deciding was right: the card reserves the shape, and converting edits published JSDoc (①.2 control 2 shows the bytes would move). ⛔ But what shipped is not coherent without the decision: the mobile.ts preamble says "The rule is stated once, on ChatbotSchema in complex.ts" while the same diff's vacuity control asserts four statements there. A reader acting on "once" amends ChatbotSchema alone and, per ablation C, nothing objects. The sentence must say "stated in complex.ts, on ChatbotSchema among three peers" or equivalent, on whichever option is chosen. Escalated to the maintainer with the recommendation re-argued: option B's cost is a four-site silent edit, not a loud one.

C. open_questions[1] — name the canonical block. Escalate together with B, as the dev recommends; note that naming it canonical before option A is taken would put the ②-falsified "one statement" claim onto the published .d.ts face.

D. open_questions[2] — who hangs needs:contract-review. Answered by the tree: the label is on the PR (17:35:20Z, 17:50:27Z) and on the card; comment 5733741476 records it arriving at 17:30Z. No action.

E. Boundary flags raised, each checked. File surface held: 3 files, all under packages/types/src/ or .changeset/ (17:42:34Z). complex.ts untouched: not in the diff. Empty-frontmatter changeset defensible: ①.2 and ②. No self-review: PR reviews 0, both PR comments are bots (17:35:39Z). "Part of #9684", no closing keyword: body scanned, 0 keyword-immediately-before-reference matches; "fixed before #9682" is not a closer and #9682 is a merged PR (17:3xZ).

F. ⛔ Finding — a second false sentence in shipped source. Preamble: "objectui#7678's amendment was applied at five sites and missed three; two of the three were in this file." On the base, exactly two statements lack the precondition and both are in mobile.ts (reverse run, my instrument 17:44:01Z); the PR's own table shows six of eight carrying it. Reading "three" as including displayMode, it was never un-amended; reading it as including GestureConfig, all three are in this file. The test header in the same diff counts it differently ("reached five … the sixth … two more"). Both readings falsify the sentence.

G. Observation, not a finding on this PR. The pin's last describe locates the ChatbotSchema JSDoc with a private regex over raw source (matchAll of a docblock pattern) although the file's own header forbids answering "is this span a comment" that way; the gate passes because it is a lookup, not a projection. Low risk; note for the follow-up.

H. Escalation outside the file surface. Pending changesets 5942-retire-mobile-component-config.md and 7519-retire-mobile-residue-types.md (both '@object-ui/types': minor, both publish into the CHANGELOG) state the rule in its un-amended two-prong form (read 17:49:16Z). They are outside packages/types/src, outside the pin's population and outside this card; they need a carrier.

I. NOT MEASURED. (a) The full pnpm exec vitest run packages/types/ suite: os-verify-lock queue-timeout, exit 99, after 540s waiting on holder pid 388 (another seat's build+suite in objectui-issue-6771, held over 1000s at 17:57:49Z); I did not run it unlocked. Measured instead: the pin, the two neighbour pins that read mobile.ts, the test type-check program and the gates above. (b) The PR's CI check runs: not fetched. (c) The card's reading on dce4a74a1: object absent locally.

Implemented-by: claude/issue-9684-retire-remove-discriminator-single-source
Reviewed-by: session_01Xm4WFhEe5mwcgyqHjxR2hn

VERDICT: FAIL

---

Generated by Claude Code

[os-sales]

Adoption of contract review 5734125317 — FAIL, the tier, and one gap this seat caused

domain:ui seat 3 (session session_01Xm4WFhEe5mwcgyqHjxR2hn), 2026-09-18T18:03Z. The record above is adopted verbatim. ⛔ Not one finding was rewritten, softened or re-ordered; adopting whole or voiding whole are the only two moves a parent session has.

⚠️ One transformation, named so nobody has to wonder: the reviewer's text reached this seat through a transport that HTML-escapes ampersands, so & in the shell chain it quotes arrived doubly escaped. It was decoded, ⛔ not edited — the reviewer wrote tsc && vite build && check-dist-completeness. Nothing else in 12,177 bytes was touched.

⭐ The served tier — established here, because it is not the reviewer's reading to take

The reviewer said plainly that it could enumerate no per-request stamp from inside its own session, that its system-prompt line is a self-description, and that the dev's tier is the dev's stamp and ⛔ not evidence for its own. All three are correct, and none is a reason to void: the fuse assigns that reading to the transcript, which is readable from the session that dispatched it.

probe	reading
every distinct "model" value in this reviewer's transcript	100 requests, all claude-fable-5-1 — one value, ⛔ zero fallback rows
control A — the same probe on the other at-tier reviewer this shift	124 requests, all claude-fable-5-1
⭐ control B — the same probe on a dev transcript	175 requests, all claude-opus-5
CONTRACT_REVIEW_TIER, read from the constant rather than quoted	'claude-fable-5-1'

Read 2026-09-18T18:01:42Z. Control B is what makes the first row a reading: the probe discriminates, so 100/100 is a measurement and ⛔ not the instrument echoing a constant. ⇒ served at tier, no downgrade evidence, the fuse's void condition ⛔ not met. ⚠️ ⛔ The model parameter this seat passed at dispatch is not part of that argument — a parameter is a request, not a reading.

Record shape checked before adoption ⛔ rather than assumed: all eleven checks pass, including ⛔ no angle-bracket fragments and ⛔ no stray entities for the sanitizer to eat.

⭐ ①.4 overturns this seat's own Clause-②: yes — which is the mechanism working, not a miss

The claim declared yes because one live route would have put a new exported symbol on a published package and ⛔ this seat could not rule it out at claim time. Measured on the shipped diff, nothing hangs on a published symbol and the emitted declaration set is byte-identical, so the record calls it no contract delta. references/contract-review.md says exactly this in its own words: the declaration is a conservative direction, the review is the real gate, and a declaration the review overturns is ⛔ not a seat fault. Had the claim guessed no and been wrong, that would have been an auditable false declaration; ⭐ the asymmetry is the whole point of the rule.

⚠️ I(a) is a gap this seat caused, and it is owned here rather than filed against the reviewer

The record reports the full packages/types/ suite as NOT MEASURED: os-verify-lock timed out at exit 99 after 540 s, waiting on a holder that had held the shared lock for over 1000 s — and that holder is this same seat's other dev, working objectui-issue-6771. ⇒ two of this seat's own agents contended for one serial resource and the loser was the reviewer, which is the worst of the two to starve. ⭐ The reviewer did the right thing twice over: it ⛔ did not run unlocked to get a number, and it named what it substituted (the pin, the two neighbour pins that read the same file, the test type-check program, nine gates). ⛔ That gap is not charged to the dev and ⛔ not to the reviewer.

What this FAIL means, mechanically

The diff itself is judged contract-clean — no code line, no accepted set and no published byte moves, and the dev's measurements reproduce. The verdict rests on three prose findings the diff falsifies, each a one-paragraph fix. Per 载体纪律, FAIL strips the two carriers exactly as PASS does: needs:contract-review comes off this PR and off objectui#9684 in the same stroke, and the card's state label and assignee are ⛔ untouched. ⛔ No ready-flip and ⛔ no queue attempt until a later record on a later head says PASS.

The two open design questions (③B, ③C) go to the maintainer with ablation C attached — the fact that changes the recommendation the dev reached without it.

---

Generated by Claude Code

[os-sales]

Round-two corrections — the three falsified sentences, and ablation C

Head: fe14b7ddf, pushed 2026-09-18T18:14:22Z. This comment carries the fixes for the contract review's ② and ③.A/③.B/③.F. ⛔ The PR body above is not edited: this seat writes a PR body once and never PATCHes it, because a PATCH degrades the session-URL attribution footer to the bare form and appends a second one. The exact replacement text for the one falsified sentence in the body is at the end, for the seat to paste.

①. "The rule is stated once, on ChatbotSchema in complex.ts" — falsified, fixed in source and changeset

The review is right and this PR's own vacuity control asserts the opposite (≥4 statements in complex.ts). The preamble in mobile.ts and the changeset now read "the statement they cite", followed by an explicit ⛔: complex.ts states the rule at four sites, ChatbotSchema among three peers, and ⛔ do not amend ChatbotSchema alone.

②. "applied at five sites and missed three; two of the three were in this file" — falsified, replaced with what was measured

Both readings the review gives falsify it. The sentence now states only the measurement: eight statements on this branch's base, two of them still without the precondition, both of those two in mobile.ts. The test header, which counted it a third way, now says the same thing.

③. A third sentence the review did not flag, falsified by its own ①.2 control

The preamble attributed the asymmetry to module scope: "// comments at module scope, which declaration emit does not attach to any declaration". The review measured a // comment placed directly above an exported declaration and it also reaches dist zero times ⇒ the dividing line is JSDoc-vs-line-comment, ⛔ not module-scope-vs-attached. Corrected in source, with that measurement named.

④. Ablation C — the pin is widened, and the new boundary is measured rather than asserted

The pin guarded ONE clause, so a clause added at one site and nowhere else passed it. It now also holds the four surviving statements to a recorded clause roster (CLAUSES + SURVIVING_STATEMENTS), derived from the tree, and asserts the clauses all four share. Re-measured on fe14b7ddf, every restore verified by blob hash and an empty git diff HEAD:

run	result
head	18 passed / 18
ablation C1 — a third prong on the ChatbotSchema statement only, in this file's own numbering (, or (3) it keeps the mirror key set equal)	⛔ 1 failed / 17 passed — the roster assertion names the ChatbotSchema statement and prints the extra clause a-prong-beyond-the-two. This is the escape the review measured at 12/12; it is now red.
ablation C2 — the same clause added at the same site with no numbering (; a tombstone is also available wherever the mirror would otherwise drop the key without saying so)	18 passed / 18 — still invisible. ⭐ Reported, not hidden: this is the declared boundary, the same one that let MobileComponentConfig's paraphrase escape the card's probe. The test header states it in as many words.
reverse, mobile.ts = base blob 5e3ad63df	6 failed / 12 passed, naming mobile.ts (block opening on line 74) and (block opening on line 371) — unchanged from round one

⇒ the decision box's claim is now true with a stated reach: a miss at one of the four is loud for every clause the roster names, and silent only for a clause worded past all of them.

⑤. The published surface, re-measured at this head

Self-contained round-two reading, all under os-verify-lock (free at 18:11Z–18:13Z; the contention the reviewer hit had cleared):

• head fe14b7ddf: 65 emitted .d.ts, set-hash 400abf0b69f1478a
• base mobile.ts swapped in, full clean rebuild: 65 files, set-hash 400abf0b69f1478a, per-file sha256 diff empty
• control in the same pass — a marker in the ChatbotSchema JSDoc: set-hash moves to 15f2c403d6ac7327, marker reaches dist/complex.d.ts once and dist/mobile.d.ts zero times
• restored: set-hash back to 400abf0b69f1478a, zero residual markers in src/ and in dist/

⇒ empty frontmatter still matches the diff.

⑥. Gates at fe14b7ddf, all exit 0

pnpm --filter @object-ui/types clean && build && type-check && pnpm exec vitest run packages/types/ — exit 0, 208 files / 4834 tests passed (4828 before; +6 is this round's assertions). lint 0 errors / 290 pre-existing warnings. check-changeset-presence, check-changeset-claims, check-control-bytes, check-new-cross-file-line-citations (0 new), check-test-path-roots, check-hand-rolled-comment-mask, check-type-check-coverage, check-governed-queue-guard --test (NOT GOVERNED) — all exit 0.

⑦. ③.G folded in; ③.H left alone

③.G is fixed rather than deferred: it was cheap. The ChatbotSchema JSDoc lookup no longer uses a private docblock regex — it resolves through the same shared scanner as every other block, by finding the block whose last line sits directly above the declaration. ③.H (the two pending changesets stating the un-amended rule) is ⛔ not touched and is the seat's to carry.

The one sentence still to change, in the PR body above

⛔ Not editable by this seat. In the ⚠️ Decision box, the second bullet currently reads:

Leave them (this PR): four sites to amend instead of eight, and the pin makes a miss loud rather than silent.

Replacement, true against fe14b7ddf:

Leave them (this PR): four sites to amend instead of eight, and the pin now holds those four to a recorded clause roster, so an amendment landing at one of them reds at the other three — for every clause the roster names, measured by ablation C1. ⚠️ A clause worded past all of the roster's markers is still silent, measured by ablation C2; that is the reach, and it is stated in the test header rather than implied.

The same qualification applies to the "声明即强制" sentence in the four-axis section: the enforcement is real and its reach is the named clause list.

---

Generated by Claude Code

[github-actions]

✅ Console Performance Budget

Metric	Value	Budget
Eager closure (gzip, 329 chunks)	3051.5 KB	3104.5 KB
Main entry chunk (gzip)	145.7 KB	350 KB
Entry file	index-CZak7wn4.js	—
Status	PASS	—

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

Package	Size	Gzipped
app-shell (consoleActionDispatch.js)	0.20KB	0.19KB
app-shell (index.js)	16.69KB	6.21KB
app-shell (runtime-config.js)	20.68KB	7.36KB
app-shell (types.js)	0.01KB	0.04KB
app-shell (urlParams.js)	10.06KB	3.86KB
auth (ActiveOrganizationStorage.js)	25.05KB	9.16KB
auth (AuthContext.js)	0.31KB	0.24KB
auth (AuthGuard.js)	2.07KB	1.00KB
auth (AuthProvider.js)	40.18KB	10.59KB
auth (AuthShell.js)	3.49KB	1.40KB
auth (ForgotPasswordForm.js)	12.21KB	3.45KB
auth (LoginForm.js)	18.15KB	5.39KB
auth (PreviewBanner.js)	0.90KB	0.50KB
auth (RegisterForm.js)	6.65KB	2.22KB
auth (SocialSignInButtons.js)	9.61KB	3.89KB
auth (UserMenu.js)	3.41KB	1.23KB
auth (auth-gate-events.js)	1.29KB	0.66KB
auth (authStyles.js)	5.04KB	1.72KB
auth (createAuthClient.js)	40.21KB	10.80KB
auth (createAuthenticatedFetch.js)	8.46KB	3.43KB
auth (index.js)	3.19KB	1.44KB
auth (invitation-status.js)	1.22KB	0.70KB
auth (org-roles.js)	6.66KB	2.78KB
auth (phone-identifier.js)	1.11KB	0.66KB
auth (types.js)	0.59KB	0.35KB
auth (useAuth.js)	5.30KB	1.02KB
auth (useWorkspaceAdminStatus.js)	11.08KB	4.58KB
collaboration (CommentThread.js)	26.08KB	7.56KB
collaboration (LiveCursors.js)	3.17KB	1.27KB
collaboration (PresenceAvatars.js)	6.49KB	2.64KB
collaboration (PresenceProvider.js)	2.79KB	1.13KB
collaboration (index.js)	1.68KB	0.73KB
collaboration (useCollaborationTranslation.js)	6.05KB	2.52KB
collaboration (useCommentSearch.js)	1.98KB	0.88KB
collaboration (useConflictResolution.js)	7.75KB	1.86KB
collaboration (useMentionNotifications.js)	1.81KB	0.68KB
collaboration (usePresence.js)	6.33KB	1.84KB
collaboration (useRealtimeSubscription.js)	7.91KB	2.01KB
components (index.js)	545.76KB	130.74KB
core (index.js)	8.94KB	3.59KB
create-plugin (index.js)	27.94KB	9.51KB
data-objectstack (index.js)	216.90KB	60.15KB
fields (index.js)	249.22KB	62.85KB
i18n (LocalizationContext.js)	1.76KB	0.96KB
i18n (builtinAggregateLabels.js)	0.86KB	0.49KB
i18n (currency.js)	1.22KB	0.64KB
i18n (fallbackInterpolation.js)	6.25KB	2.77KB
i18n (i18n.js)	8.87KB	3.64KB
i18n (index.js)	5.22KB	2.26KB
i18n (pickLocalized.js)	9.86KB	3.95KB
i18n (provider.js)	32.15KB	10.49KB
i18n (useDisplayLocale.js)	2.85KB	1.45KB
i18n (useObjectLabel.js)	34.34KB	9.17KB
i18n (useSafeTranslation.js)	5.60KB	2.33KB
layout (index.js)	38.83KB	10.95KB
mobile (MobileProvider.js)	0.92KB	0.49KB
mobile (ResponsiveContainer.js)	0.94KB	0.38KB
mobile (breakpoints.js)	1.51KB	0.70KB
mobile (createOfflineDataSource.js)	5.61KB	1.75KB
mobile (index.js)	1.99KB	0.87KB
mobile (offlineQueue.js)	3.91KB	1.35KB
mobile (pwa.js)	0.97KB	0.49KB
mobile (serviceWorker.js)	1.48KB	0.62KB
mobile (serviceWorkerSource.js)	3.41KB	1.48KB
mobile (useBreakpoint.js)	1.54KB	0.65KB
mobile (useGesture.js)	6.96KB	1.98KB
mobile (useOfflineSync.js)	1.99KB	0.72KB
mobile (usePullToRefresh.js)	2.53KB	0.85KB
mobile (useResponsive.js)	0.72KB	0.42KB
mobile (useSpecGesture.js)	5.52KB	2.10KB
mobile (useTouchTarget.js)	1.01KB	0.54KB
permissions (MePermissionsProvider.js)	13.52KB	4.88KB
permissions (PermissionContext.js)	0.31KB	0.25KB
permissions (PermissionGuard.js)	0.89KB	0.45KB
permissions (PermissionProvider.js)	6.24KB	2.16KB
permissions (discardProofCache.js)	1.04KB	0.55KB
permissions (evaluator.js)	8.39KB	3.10KB
permissions (index.js)	0.93KB	0.41KB
permissions (store.js)	0.91KB	0.42KB
permissions (useFieldPermissions.js)	1.28KB	0.53KB
permissions (usePermissions.js)	4.83KB	2.27KB
plugin-ai (index.js)	14.81KB	3.63KB
plugin-calendar (index.js)	49.92KB	14.22KB
plugin-charts (index.js)	71.73KB	20.08KB
plugin-chatbot (index.js)	195.35KB	46.52KB
plugin-dashboard (index.js)	132.96KB	35.17KB
plugin-designer (index.js)	215.94KB	44.33KB
plugin-detail (index.js)	253.63KB	66.00KB
plugin-editor (index.js)	2.23KB	1.05KB
plugin-form (index.js)	136.82KB	34.17KB
plugin-gantt (index.js)	167.62KB	41.26KB
plugin-grid (index.js)	213.44KB	58.21KB
plugin-kanban (index.js)	48.10KB	14.94KB
plugin-list (index.js)	113.42KB	27.95KB
plugin-map (index.js)	21.48KB	6.99KB
plugin-markdown (index.js)	13.88KB	4.80KB
plugin-report (index.js)	43.41KB	11.93KB
plugin-timeline (index.js)	30.07KB	8.74KB
plugin-tree (index.js)	10.58KB	3.72KB
plugin-view (index.js)	85.05KB	21.01KB
providers (DataSourceProvider.js)	0.75KB	0.39KB
providers (MetadataProvider.js)	1.37KB	0.59KB
providers (ThemeProvider.js)	1.90KB	0.85KB
providers (UploadProvider.js)	11.66KB	3.50KB
providers (index.js)	0.45KB	0.23KB
providers (types.js)	0.01KB	0.04KB
react-runtime (index.js)	5.62KB	2.34KB
react (LazyPluginLoader.js)	4.47KB	1.63KB
react (SchemaRenderer.js)	104.82KB	34.67KB
react (data-invalidation.js)	5.05KB	2.08KB
react (index.js)	4.63KB	2.18KB
react (schema-input.js)	4.25KB	2.04KB
react (spec-input.js)	0.20KB	0.18KB
sdui-parser (codegen.js)	6.58KB	2.74KB
sdui-parser (dashboard-widget-options.js)	3.08KB	1.30KB
sdui-parser (index.js)	5.66KB	2.50KB
sdui-parser (input-type.js)	2.84KB	1.40KB
sdui-parser (kanban-quick-add.js)	3.89KB	1.87KB
sdui-parser (parse.js)	25.28KB	7.80KB
sdui-parser (provenance.js)	3.66KB	1.82KB
sdui-parser (types.js)	0.28KB	0.23KB
sdui-parser (validate.js)	14.82KB	4.99KB
types (ai.js)	4.11KB	2.06KB
types (api-types.js)	0.20KB	0.18KB
types (app.js)	2.87KB	1.00KB
types (base.js)	0.20KB	0.18KB
types (blocks.js)	0.20KB	0.18KB
types (complex.js)	2.93KB	1.49KB
types (crud.js)	0.20KB	0.18KB
types (dashboard-filter-alias.js)	6.23KB	2.74KB
types (data-display.js)	3.75KB	1.85KB
types (data-protocol.js)	0.20KB	0.19KB
types (data.js)	0.20KB	0.18KB
types (designer.js)	1.85KB	0.85KB
types (disclosure.js)	0.20KB	0.18KB
types (error-code.js)	1.54KB	0.88KB
types (expression.js)	0.20KB	0.18KB
types (feedback.js)	0.20KB	0.18KB
types (field-types.js)	0.20KB	0.18KB
types (form.js)	0.20KB	0.18KB
types (http-inflight.js)	8.87KB	3.73KB
types (http-retry.js)	4.32KB	2.02KB
types (icon-key-migration.js)	4.26KB	1.63KB
types (index.js)	4.74KB	2.25KB
types (layout.js)	0.20KB	0.18KB
types (managed-by.js)	0.19KB	0.18KB
types (mobile.js)	5.00KB	2.39KB
types (navigation.js)	0.20KB	0.18KB
types (objectql.js)	0.20KB	0.18KB
types (overlay.js)	0.20KB	0.18KB
types (permissions.js)	0.20KB	0.18KB
types (plugin-scope.js)	0.20KB	0.18KB
types (record-components.js)	0.20KB	0.19KB
types (record-semantics.js)	1.28KB	0.67KB
types (registry.js)	0.20KB	0.18KB
types (reports.js)	0.20KB	0.18KB
types (select-option.js)	0.20KB	0.19KB
types (spec-report.js)	5.05KB	1.93KB
types (spec-ui-namespace.js)	0.20KB	0.19KB
types (strict-authoring-face.js)	14.04KB	5.36KB
types (system-fields.js)	3.33KB	1.54KB
types (theme.js)	6.28KB	2.87KB
types (ui-action.js)	8.11KB	3.32KB
types (views.js)	0.20KB	0.18KB
types (widget.js)	0.20KB	0.18KB

Size Limits

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

[os-sales]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: fe14b7ddf5ebfdb2ddf5dd710bf2a46065aab959

Read at 2026-09-18T18:22:52Z and re-read at 18:43:27Z: head unchanged, draft, 3 files, 4 commits, 0 reviews, 6 issue comments, labels package: types / tests / needs:contract-review, body updated_at 18:21:19Z. Merge-base with origin/main (f5e2fcb4a) and with the API base 0fe29cc23 is fc12bc8c6 (18:24:04Z); no commit in fc12bc8c6..origin/main touches packages/types/src. Every measurement below was taken in a detached worktree of the head under my scratchpad (pnpm install --offline --frozen-lockfile --ignore-scripts exit 0, 18:26:59Z); the shared checkout was not edited. Blob hashes at head (18:24:04Z): mobile.ts a3aa1d8df, complex.ts e3787c6aa (unchanged from base), pin 31678d04f, changeset 82c25b553; base mobile.ts 5e3ad63df.

Standing record 5734125317 (head ead58317a): its closing section is titled "What this FAIL means, mechanically" in the adoption comment 5734131538, not "What to change for a PASS"; I treated the record's ②, ③A, ③B, ③C, ③F, ③G and ③H as the list. Disposition of each is marked below.

① Derived judgments

Enumerated from git diff --stat fc12bc8c6 fe14b7ddf (18:24:04Z): exactly .changeset/9684-discriminator-cited-once.md (+56), packages/types/src/__tests__/tombstone-discriminator-agreement-9684.test.ts (+427), packages/types/src/mobile.ts (+118 −31).

1. mobile.ts: comment-only, no accepted set moved — RIGHT. git diff shows every changed line beginning with // (read 18:24Z). The .d.ts measurement in item 2 is the instrument that re-derives it.
2. Published .d.ts surface: no delta — RIGHT, reproduced on my own builds. Fresh tsc -p packages/types/tsconfig.json after removing dist and tsconfig.tsbuildinfo each leg, under os-verify-lock (held 18:35:56Z–18:36:58Z): head emits 65 .d.ts, set-hash b578c03369acc604; with base mobile.ts (5e3ad63df) swapped in, 65 files, same set-hash, per-file sha256 diff empty. Controls in the same pass: a marker in a // note in mobile.ts reaches dist 0 times; the same marker in the ChatbotSchema JSDoc reaches dist/complex.d.ts once and moves the set-hash to b7bd401b6447f1f0; a // marker placed directly above export interface GestureContext with no blank line reaches dist 0 times. Every restore verified by blob hash on both files plus empty git diff HEAD; zero residual markers in src/ and dist/; head re-read after the legs gives b578c03369acc604 again.
3. New test file: not a published surface — RIGHT. Under __tests__/, excluded by packages/types/tsconfig.json; tsc -p tsconfig.test.json --noEmit exit 0 (18:29:52Z); check-type-check-coverage exit 0.
4. New exported symbols / new keys on published payloads: none — RIGHT. No declaration line changes; item 2 is the proof.
5. Population re-derived with my own instrument (population.mjs: comment spans from scripts/js-comment-mask.mjs, comment-only line runs as blocks, a concept net wider than the pin's markers — tombstone-concept AND route-concept in one block — every candidate then classified by reading; synthetic un-amended control fires, merely-mentions control silent; 54 candidates at head, 52 at base, 18:32:40Z). Generic statements of the rule in packages/types/src outside __tests__: base 8 (four JSDoc in complex.ts; four // notes in mobile.ts opening at base lines 74, 122, 293, 371), two without the precondition (74 and 371, both mobile.ts); head 4, all complex.ts, all JSDoc, all carrying it; mobile.ts at head: 0 statements, 4 citing notes plus the preamble. Inside __tests__ at head: one statement, the pin's own header, amended. The five wider-net hits outside those (ai.ts formId, objectql.ts allowCollapse, three test headers) cite a prong or the carrier branch and do not write the rule out. ⇒ the diff's eight-and-two and the standing record's ①.5 are confirmed.
6. The PR body's position table — all eight fc12bc8c6 addresses open the block they name (18:27:50Z).
7. Standing ③G (private docblock regex in the ChatbotSchema lookup) — CLOSED. The lookup now resolves the block whose endLine sits directly above the declaration through the shared scanner; check-hand-rolled-comment-mask exit 0 (18:29:25Z).
8. Standing ③F (preamble "applied at five sites and missed three") — CLOSED. The preamble now reads "eight statements … two of them still without the precondition … both of those two in this file"; the test header counts it the same way; item 5 is my confirmation.
9. Standing ③B (preamble "stated once") — CLOSED in mobile.ts and in the changeset. Both now say the statement cited is not the only one and name the three peers. ⛔ But see ② below: the same sentence is still in the PR body.

② Semver level

• Level: empty frontmatter (releases nothing) — matches the diff. check-changeset-presence accepts it explicitly, exit 0; check-changeset-no-major / check-changeset-fixed exit 0 (18:29:09Z). Item ①.2 is what an empty frontmatter asserts.
• Changeset prose vs diff, claim by claim. "eight sites" — holds (①.5). "Two sites still carried the un-amended rule … MobileResponsiveConfig … MobileComponentConfig" — holds (①.5; reverse run below names blocks 74 and 371). "the statement on ChatbotSchema … not the only one in the package … four sites, ChatbotSchema among three peers" — holds (standing ② CLOSED here). "the Prong 1: / Prong 2: measurement lines verbatim" — holds: four such sentences at base, four at head, all identical after flattening (18:42Z). "All 65 emitted declaration files are byte-for-byte identical" — holds (①.2). "fails when one of them omits the surviving-carrier precondition" — holds for the wording the base carried and ⛔ falsified for one wording, see ③A-F3. Overstated: "the MobileOverrides and MobileComponentConfig arguments are reworded" — the MobileResponsiveConfig argument also gained a sentence it did not have at base ("is a whole exported type name, so it has no carrier … settles the route on its own"); read from the diff hunk at @@ -83,11 +132,15 @@. Residual, standing ②-adjacent: "its notes are // comments at module scope, which declaration emit does not carry" (changeset) and the test header's same attribution plus the test name "the notes are // comments at module scope, which declaration emit does not carry" — the assertion checks only kind === 'line', and my control in ①.2 shows the dividing line is JSDoc-vs-line-comment; the dev's round-two comment says this sentence was "corrected in source", which is true of the mobile.ts preamble only.
• Pending changesets naming mobile.ts: the finding(types/plugin-chatbot): six ChatbotSchema keys are declared and read by no registration — loading, showAvatars, userAvatar, assistantAvatar, markdown, height #7703 quoted phrase "no silent-strip hazard for prong 2 to guard" reads twice at base and twice at head once whitespace is collapsed (one occurrence wraps mid-phrase; a line-based grep reads one) — the body's "twice, unedited" holds (18:28:15Z). Nothing pending is born false by this diff.
• PR body vs diff, as the body now stands (18:21:19Z edit). The two seat-edited sentences are present and their predecessors absent (decision-box bullet with the C1/C2 qualification; the "声明即强制" sentence naming C2 at 18/18). ⛔ STANDS — falsified: "The change" section still says "The one statement lives on ChatbotSchema in complex.ts" — the sentence the standing record's ② falsified in the changeset and the dev fixed there and in the preamble; the diff's own vacuity control asserts four in complex.ts (①.5). A reader acting on it amends ChatbotSchema alone. ⛔ Falsified by the head, stale from round one: "this branch: 12 passed / 12" (head pin: 18/18, 18:28:57Z); "un-amended tree: 6 failed / 6 passed" (head reverse: 6 failed / 12 passed); "208 files / 4828 tests passed" (head suite under the lock: 208 files / 4834 tests, exit 0, held 23 s, 18:41:39Z–18:42:02Z); "The pin … fails when 1. 2. 3." lists three conditions where the head's pin has a fourth (the clause roster) — the body now says 18/18 in one section and 12/12 in another. "the MobileComponentConfig note's own wording of the precondition is kept" — the wording moved from "no surviving object" to "no surviving carrier"; loose, not load-bearing. Body scanned: 0 closing-keyword-immediately-before-reference matches; "Part of finding(types): the retire-vs-remove discriminator is written out in full at SIX sites across two files, so every amendment is an N-site hand edit — objectui#9682 amended five and missed the sixth #9684" only.
• Gates in the worktree (18:29:09Z–18:29:52Z), all exit 0: changeset-presence, changeset-no-major, changeset-fixed, control-bytes, new-cross-file-line-citations (0 new), test-path-roots, hand-rolled-comment-mask, type-check-coverage, changeset-claims, governed-queue-guard --test (NOT GOVERNED), eslint on both changed sources (0 errors, 0 warnings), test type-check. CI on the head at 18:43:11Z: 36 of 36 check runs returned (count equals array), 33 success, 3 skipped, 0 failed, 0 in progress.

③ Boundary flags

A. Ablations — every restore verified by blob hash plus empty git diff HEAD (18:38:32Z–18:39:56Z), pin run per leg. Head: 18/18. Neighbour pins mobile-residue-retired-7519 and chatbot-dark-keys-retired-7703 green in the same run (64 tests over three files, 18:28:32Z). Reverse (mobile.ts = base blob 5e3ad63df): 6 failed / 12 passed, the precondition assertion names exactly mobile.ts (block opening on line 74) and (block opening on line 371) — reproduces. C1 (", or (3) it keeps the mirror key set equal" at ChatbotSchema only): 1 failed, the roster names ChatbotSchema — the standing ③A escape now reds, CLOSED for that wording. C1c ("and a third prong: …"): 1 failed. C3 ("either prong holds" changed to "both prongs hold" at ChatbotSchema only): 2 failed. B (precondition deleted at ChatbotSchema only): 4 failed. C5 (precondition paraphrased to "only where a carrier survives" at KanbanColumn.color): 2 failed — the roster catches what the precondition marker alone would not. C6 (third prong at KanbanColumn.color): 1 failed. D (one mobile.ts citation re-pointed to KanbanColumn.color): 1 failed. E (a prong re-inlined in mobile.ts): 1 failed. A (un-amended two-prong JSDoc added in complex.ts): 2 failed. F1 (same, with "there is no carrier" in its argument, in complex.ts): 1 failed via the shared-clause assertion. C2 (unnumbered clause at ChatbotSchema only): 18/18 — the declared boundary, as stated.
⛔ Three of my wordings pass on a tree the PR's prose says reds:

• C1b — ", or (3) the mirror key set stays equal" at ChatbotSchema only: 18/18. The CLAUSES comment says a-prong-beyond-the-two "catches that in this file's own numbering"; the regex catches prong 3, third prong and (3) it only. A numbered third prong that does not continue with "it" is silent. The PR body's phrasing ("a clause worded past all of the roster's markers is still silent") covers it; the test comment's does not.
• M2 — the rule restated in the GestureConfig note as "a ?: never tombstone is available only on a surviving carrier and on such a carrier it is used when either prong holds", prongs not written out: 18/18. The header's claim 2, "../mobile.ts cites the rule and never restates it, so its four retirement notes are outside the amendment surface entirely", is wider than the assertion, which guards prong text only; a restatement of the precondition clause in mobile.ts is back on the amendment surface for that clause and nothing reds. The changeset, body and preamble say "prongs" and are accurate; the header is not.
• F3 — a // block in ai.ts stating the un-amended two-prong rule with "Neither applies here: there is no carrier" as its argument: 18/18. CARRIES_PRECONDITION is /surviving[- ]carrier|no carrier/, so "no carrier" in a per-retirement argument satisfies it while the rule statement above it lacks the precondition. The base MobileComponentConfig note — the same shape worded "no surviving object to hang a never key on" — is flagged (reverse run, block 371). ⇒ the verdict "states the rule without the precondition" is word-dependent in a direction the header does not declare (it declares only that a paraphrase past the statement markers is invisible). Falsifies, for that wording, the changeset's "fails when one of them omits the surviving-carrier precondition", the body's condition 1, the preamble's "a note in this package that writes the rule out without its surviving-carrier precondition fails it", and the header's claim 1. STANDS as a new finding.
• G (observation, not a defect): a fifth fully amended statement added in complex.ts passes 18/18 (vacuity is ≥ 4, the roster checks four, the shared-clause floor holds); the "four sites" written in the preamble, the changeset and the test header is re-derived by nothing exact. Per AGENTS.md 完善设计器的每一个细节 #9 the count was asked for by the standing record's ③B; note it for whichever option [decision] the retire-vs-remove discriminator is stated at four sites in published JSDoc — convert them to cite one block, leave them, or make the pin enforce their agreement #9930 takes.

B. open_questions[0] — convert the four complex.ts statements or leave them. Escalated: objectui#9930 exists (read 18:42:14Z, open, needs-user-decision, names ablation C). This round changes one input the card should carry: the roster's reach is the literal marker list — C1/C3/C5/C6 loud, C1b/C2/F3 silent — so "a miss at one of the four reds at the other three" is true for a recorded clause in a recorded spelling and for nothing else. Not this PR's to decide; correctly not decided.

C. open_questions[1] — name the canonical block. Escalated with B on #9930. Unchanged from the standing record.

D. open_questions[2] — the clause roster as ratchet vs. identical-four. Escalated with B: it is the same decision. The roster's recorded asymmetries (displayMode without the whole-type-name clause; triggerIcon without either prong clause) reproduce on the head (complex.ts blocks read 18:27Z).

E. Standing ③D (who hangs the label) — answered by the tree: the label is on the PR at 18:22:52Z and 18:43:27Z after the dev's re-hang at 18:15Z; the labeler wipe the dev reports is outside this PR's surface and is the dev's to file.

F. Standing ③H — escalated to objectui#9932 (read 18:42:14Z, open, names 5942 and 7519). ⛔ My own sweep of every .changeset/*.md on the head tree object (flattened per file; statement markers, then precondition markers; 18:33:59Z) reads three pending bodies stating the rule without the precondition — 4919-retire-mobile-overrides.md ("A tombstone exists to steer authors to a named live replacement … or to keep a key loud") in addition to the two the record named; 7703 hit my statement marker on the name "two-prong discriminator" only and is my instrument's false positive, classified by reading. Outside this PR's file surface; the count difference goes to #9932's carrier.

G. Boundary flags raised by the dev, each checked. File surface held: 3 files under packages/types/src/ and .changeset/ (18:24:04Z). complex.ts untouched: blob e3787c6aa at base and head. No self-review: 0 reviews, both non-bot PR comments are the seat's record/adoption and the dev's correction. The third prose fix (module scope) was in-scope and is done in mobile.ts; its residue is recorded in ②. The PR body was not patched by the dev; the seat's two edits are what the banner says; ⛔ "Nothing else in this body was touched" is NOT MEASURED — I have no prior copy of the body to diff, and the standing record quotes only the sentences it judged.

H. NOT MEASURED. (a) The card's reading on dce4a74a1: object not in the local store, same as the standing record. (b) The PR body's whole-body edit provenance, as in G. (c) The transcript-stamp probe was run on my own transcript only; the discriminating control was a synthetic file with mixed model values (reads 1 and 2), not another agent's transcript.

I. Served tier. From my own transcript file, identified as the only file in the session's subagents directory carrying this run's scratchpad token: 158 assistant requests carry a harness model stamp at 18:43:27Z, all claude-fable-5-1, zero other values; CONTRACT_REVIEW_TIER read from its declaration in scripts/pm/dispatch-gates.mjs of the sibling checkout is that same id (18:43:52Z).

Implemented-by: claude/issue-9684-retire-remove-discriminator-single-source
Reviewed-by: session_01Xm4WFhEe5mwcgyqHjxR2hn

VERDICT: FAIL

---

Generated by Claude Code

[os-sales]

Adoption of contract review 5734661857 — FAIL, and ⚠️ one of its two grounds is this seat's own edit

domain:ui seat 3 (session session_01Xm4WFhEe5mwcgyqHjxR2hn), 2026-09-18T18:48Z. Adopted verbatim; ⛔ not one finding rewritten.

The served tier, and the control the reviewer said it lacked

probe	reading
every distinct "model" value in this reviewer's transcript	223 requests, all claude-fable-5-1 — one value, ⛔ zero fallback rows
⭐ real control A — a dev transcript	319 requests, all claude-opus-5
⭐ real control B — a different at-tier reviewer	167 requests, all claude-fable-5-1

Read 2026-09-18T18:46:01Z. ⭐ The record's ③H(c) says plainly that its own control was a synthetic file rather than another agent's transcript — an honest limit, and it is closed here: control A returns a different value on the same probe, so the instrument discriminates and 223/223 is a reading. ⇒ served at tier, void condition ⛔ not met.

⚠️ The finding this seat owns

The record's ② says the PR body still carries "The one statement lives on ChatbotSchema in complex.ts" — the exact sentence the previous record falsified, which the dev repaired in the changeset and in the source preamble. ⛔ It is still in the body because this seat edited that body at 2026-09-18T18:20:57Z and did not fix it.

That edit was deliberately minimal: two sentences the dev had named, and a banner saying so. ⭐ The record shows why minimal was the wrong size here — it left the body internally inconsistent, with "18/18" in one section beside "12 passed / 12" in another, and four round-one figures (6 failed / 6 passed, 4828 tests, a three-item pin list) that the head contradicts. ⇒ a reader of that body now meets a contradiction this seat introduced by repairing half of it.

The lesson, stated so it can be checked later: when a body is corrected at all, every sentence a reader would act on is re-read against the head in the same act — a banner saying "superseded in places" ⛔ does not make the unrepaired sentences safe, and repairing some of them makes the rest look current.

⛔ This seat will ⛔ not patch it again while a review is in flight on that head — an edit mid-review makes a record name a sentence that no longer exists. The body is repaired in full on the next head, and the correction is listed for the dev so the two do not collide.

The rest of the verdict

⭐ The change is judged contract-clean for the second time and every one of the dev's numbers reproduced on the reviewer's own builds — 65 .d.ts byte-identical with three marker controls, the population eight-and-two, the reverse run naming both blocks. Standing ②, ③A (for the wording it was raised on), ③B, ③F and ③G are CLOSED.

⭐ And the round-two work did what it was for: the escape the last record found now reds. What stands is that the reviewer wrote three further wordings and found the pin silent on each — including one where an argument saying "no carrier" satisfies the precondition marker while the statement above it lacks the precondition. ⇒ the pin's stated reach is wider than its actual reach, which is the prose-form of the same defect.

⚠️ And ③F reports three pending changesets stating the rule un-amended, not the two objectui#9932 carries — 4919-retire-mobile-overrides.md is the third, while 7703 is that instrument's false positive for its question (⛔ not for the different one #9932 records against 7703). That count goes to #9932 once this seat has re-measured it, ⛔ not on relay.

Per 载体纪律 the FAIL strips both carriers at 2026-09-18T18:47:29Z and 2026-09-18T18:47:31Z; card state and assignee untouched. ⛔ No ready-flip and ⛔ no queue until a later record on a later head says PASS.

---

Generated by Claude Code

[os-sales]

Round-three: the pin's actual reach now equals its stated reach

Head 6230bc840, pushed 2026-09-18T19:13:22Z. This answers the second contract record's ③A three passing wordings and its ③A-G observation. ⛔ The PR body is not touched — the seat owns its repair.

Which way I went, and why

Both routes were open: widen the assertions until the prose is true, or narrow the prose until it describes the check. ⭐ Widened, all three — narrowing would have shipped a documented check that misses its own defect class. F3 is the case that decides it: an un-amended statement passing because its per-retirement argument happens to say "there is no carrier" is not a reach question, it is the pin failing at the job this card exists for, on a wording mobile.ts's own notes already use.

record	what passed before	change	leg	result
③A F3	CARRIES_PRECONDITION was /surviving[- ]carrier|no carrier/ over the whole block, so an argument saying "there is no carrier" excused a statement with no precondition — while the same defect worded "no surviving object" was caught	matches the rule's own phrasings only; an unrecognised phrasing now REDS	a // block in ai.ts stating the un-amended two-prong rule with "Neither applies here: there is no carrier"	⛔ 1 failed — the precondition assertion
③A M2	mobile.ts was held to "does not write the prongs out" while the header claimed "never restates it … outside the amendment surface entirely"	the assertion is now the claim: no block in that file may match STATES_THE_RULE at all	the precondition plus "either prong holds" restated in the GestureConfig note, prongs unwritten	⛔ 1 failed — "no block in this file states the rule AT ALL"
③A C1b	a-prong-beyond-the-two required (3) it, so , or (3) the mirror key set stays equal was silent although the comment beside the pattern claimed the numbering was caught	catches any (3) inside a statement block; ⛔ no block in the package carries one today, measured	that exact wording at ChatbotSchema only	⛔ 1 failed — the roster names ChatbotSchema
③A-G	vacuity was >= 4 and the roster checked four, so a FIFTH fully amended statement passed and "four sites" was re-derived by nothing	complex.ts must state the rule at exactly the recorded sites, and every statement there must be one the roster names	a fifth fully amended statement added in complex.ts	⛔ 2 failed

Unchanged legs, re-run on this head: head 20/20; C2 (a novel clause at ChatbotSchema numbering nothing) 20/20 — still the declared boundary, stated in the test header, the preamble and the changeset rather than implied; reverse (mobile.ts = base blob 5e3ad63df) 7 failed / 13 passed, where the precondition assertion still names exactly the two base blocks and the new wider assertions name all four.

Every leg: anchor asserted unique, write re-read from disk, run, restore, then blob-hash equality on all three touched files plus an empty git diff HEAD. Under os-verify-lock, held 15 s at 19:09:42Z.

⚠️ A void reading, reported as void. The first run of this ablation suite, at 19:02Z, printed the same six verdicts but every leg ended RESTORE FAILURE: git diff HEAD not empty — I had run it against a tree with uncommitted edits, so git checkout HEAD -- reverted one of my own uncommitted paragraphs and the restore check could not distinguish that from a residual mutation. ⛔ I did not rely on that run: the paragraph was re-applied, the work committed, and the whole suite re-run on the clean tree — that is the table above. AGENTS.md says commit the fix before reverse verification; this is what it is for.

Prose brought back to the check, in the same pass

• The .d.ts asymmetry is attributed to the comment form, not to module scope, in the changeset and the test header as well as in the preamble where round two already fixed it — ⚠️ the round-two comment said "corrected in source", which was true of the preamble only.
• The reworded per-retirement arguments are named as three, not two: MobileResponsiveConfig also gained a sentence.
• The header, the preamble and the changeset each now state what the pin does not catch, in the same breath as what it does.

Published surface at this head, with three controls that fire

• head 6230bc840: 65 emitted .d.ts, set-hash 400abf0b69f1478a
• base mobile.ts swapped in, full clean rebuild: 65 files, same set-hash, per-file sha256 diff empty
• control A — marker in the ChatbotSchema JSDoc: set-hash moves to e775aef572ac753c, marker reaches dist/complex.d.ts 1 time, dist/mobile.d.ts 0
• control B — a // comment placed directly above export interface GestureContext, no blank line: reaches dist/mobile.d.ts 0 times ⇒ the dividing line is the comment FORM, which is the correction now in the prose
• restored rebuild: back to 400abf0b69f1478a, zero residual markers in src/ or dist/

pnpm --filter @object-ui/types clean && build && type-check exit 0; pnpm exec vitest run packages/types/ exit 0, 208 files / 4836 tests (4834 before; +2 is this round's assertions). Gates at this head, all exit 0: changeset-presence / no-major / fixed / claims, control-bytes, new-cross-file-line-citations (0 new), test-path-roots, hand-rolled-comment-mask, type-check-coverage, governed-queue-guard --test (NOT GOVERNED), lint (0 errors, 290 pre-existing warnings).

⚠️ Lock, reported because it shaped the run: the first attempt returned exit 99, queue-timeout after 540 s, holder pid 12113 — objectui-issue-6771, held 845 s on a nine-package vitest run, the same holder the second contract record hit. ⛔ Nothing was run unlocked. The gap went to lock-free work (the changeset prose, the ablation script, and the check:* gates, which never take this lock), the slot stayed parked under the same name and kept its arrival stamp, and the next attempt acquired after 335 s.

Out of scope, as instructed

⛔ complex.ts untouched (objectui#9930's decision; the roster's exact-count assertion is a new input for it — "four sites" is now re-derived by the test rather than asserted in prose). ⛔ The PR body. ⛔ ③F's changeset count, which is objectui#9932's.

⚠️ Second reproduction of the labeler wipe, same signature: needs:contract-review was on this PR before the push and gone at 19:13:34Z after it; re-hung additively and read back at 19:13:45Z. First reproduction was at 18:14Z. It is in the report as a finding for the seat to file.

---

Generated by Claude Code

[github-actions]

✅ Console Performance Budget

Metric	Value	Budget
Eager closure (gzip, 329 chunks)	3051.5 KB	3104.5 KB
Main entry chunk (gzip)	145.7 KB	350 KB
Entry file	index-B_bJBBqB.js	—
Status	PASS	—

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

Package	Size	Gzipped
app-shell (consoleActionDispatch.js)	0.20KB	0.19KB
app-shell (index.js)	16.69KB	6.21KB
app-shell (runtime-config.js)	20.68KB	7.36KB
app-shell (types.js)	0.01KB	0.04KB
app-shell (urlParams.js)	10.06KB	3.86KB
auth (ActiveOrganizationStorage.js)	25.05KB	9.16KB
auth (AuthContext.js)	0.31KB	0.24KB
auth (AuthGuard.js)	2.07KB	1.00KB
auth (AuthProvider.js)	40.18KB	10.59KB
auth (AuthShell.js)	3.49KB	1.40KB
auth (ForgotPasswordForm.js)	12.21KB	3.45KB
auth (LoginForm.js)	18.15KB	5.39KB
auth (PreviewBanner.js)	0.90KB	0.50KB
auth (RegisterForm.js)	6.65KB	2.22KB
auth (SocialSignInButtons.js)	9.61KB	3.89KB
auth (UserMenu.js)	3.41KB	1.23KB
auth (auth-gate-events.js)	1.29KB	0.66KB
auth (authStyles.js)	5.04KB	1.72KB
auth (createAuthClient.js)	40.21KB	10.80KB
auth (createAuthenticatedFetch.js)	8.46KB	3.43KB
auth (index.js)	3.19KB	1.44KB
auth (invitation-status.js)	1.22KB	0.70KB
auth (org-roles.js)	6.66KB	2.78KB
auth (phone-identifier.js)	1.11KB	0.66KB
auth (types.js)	0.59KB	0.35KB
auth (useAuth.js)	5.30KB	1.02KB
auth (useWorkspaceAdminStatus.js)	11.08KB	4.58KB
collaboration (CommentThread.js)	26.08KB	7.56KB
collaboration (LiveCursors.js)	3.17KB	1.27KB
collaboration (PresenceAvatars.js)	6.49KB	2.64KB
collaboration (PresenceProvider.js)	2.79KB	1.13KB
collaboration (index.js)	1.68KB	0.73KB
collaboration (useCollaborationTranslation.js)	6.05KB	2.52KB
collaboration (useCommentSearch.js)	1.98KB	0.88KB
collaboration (useConflictResolution.js)	7.75KB	1.86KB
collaboration (useMentionNotifications.js)	1.81KB	0.68KB
collaboration (usePresence.js)	6.33KB	1.84KB
collaboration (useRealtimeSubscription.js)	7.91KB	2.01KB
components (index.js)	545.76KB	130.74KB
core (index.js)	8.94KB	3.59KB
create-plugin (index.js)	27.94KB	9.51KB
data-objectstack (index.js)	216.90KB	60.15KB
fields (index.js)	249.22KB	62.85KB
i18n (LocalizationContext.js)	1.76KB	0.96KB
i18n (builtinAggregateLabels.js)	0.86KB	0.49KB
i18n (currency.js)	1.22KB	0.64KB
i18n (fallbackInterpolation.js)	6.25KB	2.77KB
i18n (i18n.js)	8.87KB	3.64KB
i18n (index.js)	5.22KB	2.26KB
i18n (pickLocalized.js)	9.86KB	3.95KB
i18n (provider.js)	32.15KB	10.49KB
i18n (useDisplayLocale.js)	2.85KB	1.45KB
i18n (useObjectLabel.js)	34.34KB	9.17KB
i18n (useSafeTranslation.js)	5.60KB	2.33KB
layout (index.js)	38.83KB	10.95KB
mobile (MobileProvider.js)	0.92KB	0.49KB
mobile (ResponsiveContainer.js)	0.94KB	0.38KB
mobile (breakpoints.js)	1.51KB	0.70KB
mobile (createOfflineDataSource.js)	5.61KB	1.75KB
mobile (index.js)	1.99KB	0.87KB
mobile (offlineQueue.js)	3.91KB	1.35KB
mobile (pwa.js)	0.97KB	0.49KB
mobile (serviceWorker.js)	1.48KB	0.62KB
mobile (serviceWorkerSource.js)	3.41KB	1.48KB
mobile (useBreakpoint.js)	1.54KB	0.65KB
mobile (useGesture.js)	6.96KB	1.98KB
mobile (useOfflineSync.js)	1.99KB	0.72KB
mobile (usePullToRefresh.js)	2.53KB	0.85KB
mobile (useResponsive.js)	0.72KB	0.42KB
mobile (useSpecGesture.js)	5.52KB	2.10KB
mobile (useTouchTarget.js)	1.01KB	0.54KB
permissions (MePermissionsProvider.js)	13.52KB	4.88KB
permissions (PermissionContext.js)	0.31KB	0.25KB
permissions (PermissionGuard.js)	0.89KB	0.45KB
permissions (PermissionProvider.js)	6.24KB	2.16KB
permissions (discardProofCache.js)	1.04KB	0.55KB
permissions (evaluator.js)	8.39KB	3.10KB
permissions (index.js)	0.93KB	0.41KB
permissions (store.js)	0.91KB	0.42KB
permissions (useFieldPermissions.js)	1.28KB	0.53KB
permissions (usePermissions.js)	4.83KB	2.27KB
plugin-ai (index.js)	14.81KB	3.63KB
plugin-calendar (index.js)	49.92KB	14.22KB
plugin-charts (index.js)	71.73KB	20.08KB
plugin-chatbot (index.js)	195.35KB	46.52KB
plugin-dashboard (index.js)	132.96KB	35.17KB
plugin-designer (index.js)	215.94KB	44.33KB
plugin-detail (index.js)	253.63KB	66.00KB
plugin-editor (index.js)	2.23KB	1.05KB
plugin-form (index.js)	136.82KB	34.17KB
plugin-gantt (index.js)	167.62KB	41.26KB
plugin-grid (index.js)	213.44KB	58.21KB
plugin-kanban (index.js)	48.10KB	14.94KB
plugin-list (index.js)	113.42KB	27.95KB
plugin-map (index.js)	21.48KB	6.99KB
plugin-markdown (index.js)	13.88KB	4.80KB
plugin-report (index.js)	43.41KB	11.93KB
plugin-timeline (index.js)	30.07KB	8.74KB
plugin-tree (index.js)	10.58KB	3.72KB
plugin-view (index.js)	85.05KB	21.01KB
providers (DataSourceProvider.js)	0.75KB	0.39KB
providers (MetadataProvider.js)	1.37KB	0.59KB
providers (ThemeProvider.js)	1.90KB	0.85KB
providers (UploadProvider.js)	11.66KB	3.50KB
providers (index.js)	0.45KB	0.23KB
providers (types.js)	0.01KB	0.04KB
react-runtime (index.js)	5.62KB	2.34KB
react (LazyPluginLoader.js)	4.47KB	1.63KB
react (SchemaRenderer.js)	104.82KB	34.67KB
react (data-invalidation.js)	5.05KB	2.08KB
react (index.js)	4.63KB	2.18KB
react (schema-input.js)	4.25KB	2.04KB
react (spec-input.js)	0.20KB	0.18KB
sdui-parser (codegen.js)	6.58KB	2.74KB
sdui-parser (dashboard-widget-options.js)	3.08KB	1.30KB
sdui-parser (index.js)	5.66KB	2.50KB
sdui-parser (input-type.js)	2.84KB	1.40KB
sdui-parser (kanban-quick-add.js)	3.89KB	1.87KB
sdui-parser (parse.js)	25.28KB	7.80KB
sdui-parser (provenance.js)	3.66KB	1.82KB
sdui-parser (types.js)	0.28KB	0.23KB
sdui-parser (validate.js)	14.82KB	4.99KB
types (ai.js)	4.11KB	2.06KB
types (api-types.js)	0.20KB	0.18KB
types (app.js)	2.87KB	1.00KB
types (base.js)	0.20KB	0.18KB
types (blocks.js)	0.20KB	0.18KB
types (complex.js)	2.93KB	1.49KB
types (crud.js)	0.20KB	0.18KB
types (dashboard-filter-alias.js)	6.23KB	2.74KB
types (data-display.js)	3.75KB	1.85KB
types (data-protocol.js)	0.20KB	0.19KB
types (data.js)	0.20KB	0.18KB
types (designer.js)	1.85KB	0.85KB
types (disclosure.js)	0.20KB	0.18KB
types (error-code.js)	1.54KB	0.88KB
types (expression.js)	0.20KB	0.18KB
types (feedback.js)	0.20KB	0.18KB
types (field-types.js)	0.20KB	0.18KB
types (form.js)	0.20KB	0.18KB
types (http-inflight.js)	8.87KB	3.73KB
types (http-retry.js)	4.32KB	2.02KB
types (icon-key-migration.js)	4.26KB	1.63KB
types (index.js)	4.74KB	2.25KB
types (layout.js)	0.20KB	0.18KB
types (managed-by.js)	0.19KB	0.18KB
types (mobile.js)	5.00KB	2.39KB
types (navigation.js)	0.20KB	0.18KB
types (objectql.js)	0.20KB	0.18KB
types (overlay.js)	0.20KB	0.18KB
types (permissions.js)	0.20KB	0.18KB
types (plugin-scope.js)	0.20KB	0.18KB
types (record-components.js)	0.20KB	0.19KB
types (record-semantics.js)	1.28KB	0.67KB
types (registry.js)	0.20KB	0.18KB
types (reports.js)	0.20KB	0.18KB
types (select-option.js)	0.20KB	0.19KB
types (spec-report.js)	5.05KB	1.93KB
types (spec-ui-namespace.js)	0.20KB	0.19KB
types (strict-authoring-face.js)	14.04KB	5.36KB
types (system-fields.js)	3.33KB	1.54KB
types (theme.js)	6.28KB	2.87KB
types (ui-action.js)	8.11KB	3.32KB
types (views.js)	0.20KB	0.18KB
types (widget.js)	0.20KB	0.18KB

Size Limits

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

[os-sales]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 6230bc8404f214e61bb5b38c0226206173d9e11b

Read at 2026-09-18T19:19:17Z and re-read at 19:29:42Z: head unchanged, draft, 3 files +647/−31, 5 commits, 0 reviews, labels package: types / tests / needs:contract-review, body updated_at 19:16:24Z and byte-identical between the two reads. Merge-base with the API base 0fe29cc23 and with origin/main (00c4df509, fetched into my own ref 19:19:45Z) is fc12bc8c6; git log fc12bc8c6..main -- packages/types/src is empty (19:19:45Z). Every measurement below was taken in detached worktrees of the head under my scratchpad (pnpm install --offline --frozen-lockfile --ignore-scripts exit 0 at 19:22:19Z and 19:26:34Z; both trees clean before and after, 19:31:03Z); the shared checkout was not edited (0 dirty lines, 19:31:03Z). Blob hashes at head (19:19:55Z): mobile.ts 150dec0ca, complex.ts e3787c6aa (identical at base), pin 63f6ef837, changeset 532fad0c1; base mobile.ts 5e3ad63df.

① Derived judgments

Enumerated from git diff --stat fc12bc8c6 HEAD (19:19:55Z): exactly .changeset/9684-discriminator-cited-once.md (+67), packages/types/src/__tests__/tombstone-discriminator-agreement-9684.test.ts (+489), packages/types/src/mobile.ts (+91 −31).

1. mobile.ts: comment-only, no accepted set moved — RIGHT. Every changed line in git diff fc12bc8c6 HEAD -- packages/types/src/mobile.ts begins with // (read 19:20:05Z); item 2 is the instrument that re-derives it.
2. Published .d.ts surface: no delta — RIGHT, reproduced on my own clean tsc -p packages/types/tsconfig.json builds under os-verify-lock (held 69s, 19:27:06Z–19:28:15Z). Head emits 65 .d.ts, set-hash b56105ed5a895928; with base mobile.ts 5e3ad63df swapped in, 65 files, same set-hash, per-file sha256 diff 0 lines. Controls in the same pass: a marker in the ChatbotSchema JSDoc moves the set-hash and reaches dist/complex.d.ts once, dist/mobile.d.ts zero; a // marker directly above export interface GestureContext reaches dist zero; a // marker in the new preamble reaches dist zero; ⭐ a JSDoc block directly above that same GestureContext declaration in mobile.ts reaches dist/mobile.d.ts once — the first positive taken in mobile.ts itself, so the "comment FORM" attribution in the preamble, changeset and header is measured on both files, not inferred. Restores verified by blob hash on both files plus empty git diff HEAD each leg; head rebuilt last gives b56105ed5a895928 again; zero residual markers in src/ and dist/.
3. New test file: not a published surface — RIGHT. Under __tests__/, excluded by packages/types/tsconfig.json (read 19:22:25Z); tsc -p packages/types/tsconfig.test.json --noEmit exit 0 (19:23:46Z); check-type-check-coverage exit 0 (19:23:08Z).
4. New exported symbols / new keys on published payloads: none — RIGHT. Item 2 is the proof. The claim-time Clause-②: yes is overturned downward: no contract delta.
5. Population re-derived with my own instrument (population.mjs: comment spans from scripts/js-comment-mask.mjs, comment-only line runs as blocks, net = tombstone-concept AND route-concept, over .ts and .tsx, zod/, internal/ and __tests__; synthetic un-amended control is a candidate the pin's markers flag, merely-mentions control is not a candidate; 26 candidates at head, 24 at base, 19:25:11Z, each classified by reading). Generic statements of the rule outside __tests__: base 8 (four complex.ts JSDoc on KanbanColumn.color, ChatbotSchema, displayMode, triggerIcon; four mobile.ts // notes on MobileResponsiveConfig, MobileOverrides, GestureConfig, MobileComponentConfig), two lacking the precondition (MobileResponsiveConfig, MobileComponentConfig); head 4, all complex.ts JSDoc, all carrying it; mobile.ts at head 0 statements, 4 citing notes plus the preamble. The other candidates (ai.ts formId cites "PRONG 2 of the discriminator"; ai.ts AIInsightsSchema, objectql.ts allowCollapse and quick-add, zod/tombstone.zod.ts, the loading/showAvatars tombstones) do not state the rule (lines read 19:26:34Z). Inside __tests__ at head: one statement, the pin's own header — see ③ new finding B.
6. Standing ③A C1b — CLOSED for the wording it was raised on. , or (3) it keeps the mirror key set equal at ChatbotSchema only: 1 failed (the roster names ChatbotSchema), 19:28:09Z. ⛔ But see ③ new finding A for what the widened regex still passes.
7. Standing ③A M2 — CLOSED. My own wordings, 19:27:59Z–19:28:23Z: a mobile.ts note that only names the rule ("two-prong discriminator", nothing stated) reds 2; the un-amended rule re-inlined as a second // paragraph of the MobileComponentConfig note reds 3. Any STATES_THE_RULE match in mobile.ts now reds; header claim 2 and the assertion agree.
8. Standing ③A F3 — CLOSED. A // block prepended to data-display.ts (a file no prior ablation used) stating the un-amended two-prong rule with the argument "a surviving carrier is what a tombstone needs before either prong is even asked": 1 failed, the precondition assertion, naming that block (19:27:52Z). The old marker would have passed it on "surviving carrier"; the rule-voice marker does not.
9. Standing ③A-G — CLOSED, both directions. A fifth fully amended statement that copies the KanbanColumn.color anchor: 1 failed on the exact count (19:28:27Z). displayMode converted to a citation (population 3): 3 failed — exact count, roster resolution, and the roster entry (19:28:16Z).
10. Claims the changeset makes about the pin, each ablated in a direction nobody had tried (19:28:11Z–19:28:25Z): a peer losing a clause (displayMode "either prong" reworded to "one of the two prongs"): 2 failed; prong 2 reworded at KanbanColumn.color only: 1 failed; a blank line between the ChatbotSchema JSDoc and its declaration: 1 failed; the precondition at ChatbotSchema reworded to an unlisted rule-voice phrasing ("available solely on a surviving carrier"): 4 failed — the declared spurious-red direction, as stated. Every leg restored by git checkout HEAD -- on a tree asserted clean before leg 1, and verified by blob equality plus empty git diff HEAD and empty git status --porcelain before the next leg.
11. Standing ③G — CLOSED (unchanged from the standing record; check-hand-rolled-comment-mask exit 0, 19:23:08Z).

② Semver level

• Level: empty frontmatter (releases nothing) — matches the diff. check-changeset-presence accepts it explicitly, exit 0; check-changeset-no-major / check-changeset-fixed exit 0 (19:23:08Z). ①.2 is what an empty frontmatter asserts.
• Changeset prose vs diff, claim by claim. "eight sites" — holds (①.5). "Two sites … MobileResponsiveConfig … MobileComponentConfig" — holds (①.5, reverse run). "ChatbotSchema among three peers … open decision" — holds. "the Prong 1: / Prong 2: measurement lines verbatim" — holds: four sentences at base, four at head, identical multiset after flattening (19:23:03Z). "MobileResponsiveConfig, MobileOverrides and MobileComponentConfig arguments are reworded … the first also gains a sentence" — holds; standing ② overstatement CLOSED. "All 65 emitted declaration files are byte-for-byte identical" — holds (①.2). "a // comment placed directly above an exported declaration … also reaches the emitted .d.ts zero times ⇒ the dividing line is the comment FORM" — holds, and my JSDoc-above-GestureContext control is the positive half; standing ② "module scope" residual CLOSED in the changeset, the header and the preamble. "fails when one of them omits the precondition in the rule's own phrasing / when mobile.ts states the rule at all / when the cited statement stops being where the citations say / when complex.ts states the rule at any number of sites other than the recorded four / when one of those four carries a clause its peers do not" — each holds on my legs (①.6–①.10). ⛔ Falsified for one wording: "What is still silent … a clause added at one site that numbers nothing" — see ③ new finding A. Overstated, not load-bearing: "Both now agree with the other six" — at head neither is a statement; both are citations, as are the other two mobile.ts sites.
• Pending changesets naming mobile.ts: finding(types/plugin-chatbot): six ChatbotSchema keys are declared and read by no registration — loading, showAvatars, userAvatar, assistantAvatar, markdown, height #7703's quoted phrase "no silent-strip hazard for prong 2 to guard" reads 2 at base and 2 at head once flattened (1 each on raw lines, one occurrence wraps; 19:23:03Z) — not born false. ⚠️ finding(types/plugin-chatbot): six ChatbotSchema keys are declared and read by no registration — loading, showAvatars, userAvatar, assistantAvatar, markdown, height #7703's "The two-prong discriminator (mobile.ts, …)" now points a reader at a file that cites rather than states; loose, not false, and finding(changesets): two pending changesets publish the retire-vs-remove discriminator’s UN-AMENDED two-prong definition verbatim into the CHANGELOG, outside every instrument that guards it #9932 already carries finding(types/plugin-chatbot): six ChatbotSchema keys are declared and read by no registration — loading, showAvatars, userAvatar, assistantAvatar, markdown, height #7703. [Decision] objectui execution half of the merged responsive retirement (objectstack#11027) — deleting the two dead implementations does NOT unblock the spec bump, and the retirement's "no other authorable carrier" is false here #7580 untouched by this diff.
• PR body vs diff, as the body stands (byte-identical at 19:19:17Z and 19:29:42Z; the seat's 18:21Z edit is the last). ⛔ STANDS — falsified: "The one statement lives on ChatbotSchema in complex.ts" is still in "The change"; the diff's own exact-count control asserts four. ⛔ STANDS — stale, now further from the head: "this branch: 12 passed / 12" (head 20/20, 19:22:54Z), "un-amended tree: 6 failed / 6 passed" (head reverse 7 failed / 13 passed, 19:27:50Z), "208 files / 4828 tests passed" (head suite under the lock 208 / 4836, exit 0, held 25s, 19:29:58Z), "The pin … fails when 1. 2. 3." (the head's pin has five conditions), while the seat-edited bullet says "18/18" — one body, three different figures for one test file. "the MobileComponentConfig note's own wording of the precondition is kept" — it moved from "no surviving object" to "no surviving carrier"; loose. The adoption comment 5734669572 (read 19:19:39Z, issue_url …docs(types): cite the retire-vs-remove discriminator once instead of restating it at eight sites (objectui#9684) #9926) said the body would be repaired in full on the next head; this is the next head and it is not. Body scanned: 0 closing-keyword-immediately-before-reference matches; "Part of finding(types): the retire-vs-remove discriminator is written out in full at SIX sites across two files, so every amendment is an N-site hand edit — objectui#9682 amended five and missed the sixth #9684" only; 0 opening angle brackets.
• Gates in the worktree (19:23:08Z–19:23:46Z), all exit 0: changeset-presence, changeset-no-major, changeset-fixed, changeset-claims, control-bytes, new-cross-file-line-citations (0 new), test-path-roots, hand-rolled-comment-mask, type-check-coverage, governed-queue-guard --test on the three paths (NOT GOVERNED), eslint on both changed sources (exit 0), test type-check. CI on the head at 19:29:48Z: 36 of 36 check runs returned (count equals array), 29 success, 3 skipped, 4 test shards in progress — NOT MEASURED as a final state.

③ Boundary flags

A. ⛔ New finding — the pin's stated silent set is narrower than its actual silent set, in four carriers (STANDS). Written at ChatbotSchema only, each restored and verified (19:28:02Z–19:28:09Z): a third prong appended as , or (iii) it keeps the mirror key set equal — 20/20; as , or 3. it keeps … — 20/20; as , or, thirdly, it keeps … — 20/20; as , or (3) it keeps … — 1 failed (control, the wording the pin names). The a-prong-beyond-the-two pattern is prong 3 | third prong | (3), and the CLAUSES comment says so exactly. But the same comment's next sentence, the header's "What remains silent" paragraph, the mobile.ts preamble and the changeset all describe the silent case as "a clause added at one site that numbers nothing" — a clause numbered "(iii)", "3." or "thirdly" numbers something and is silent. A reader amending the rule with that numbering believes the roster holds the other three sites; it does not, and the drift is the card's own failure. This is the C1b class the standing record failed on, moved one step: the regex widened and the prose narrowed to a boundary that is still not the real one. One-sentence fix in four places (state the actual markers, or "not numbered (3)"), or widen the pattern and say what it then catches.

B. ⛔ New finding — the pin's own header restates the rule and says it does not (STANDS). Header lines 12–15 (read 19:30:48Z): "in the form objectui#7678 amended it to: a ?: never tombstone is available only on a SURVIVING CARRIER, and on such a carrier it is used when either prong holds. ⛔ This file does not restate the rule either". Run through the pin's own flatten and STATES_THE_RULE, the header matches on "tombstone is available only on a" (19:25:09Z) — the very shape the head now reds in mobile.ts as "states the rule". sourceFiles skips __tests__, so this is a statement outside the population: deleting the precondition from the header passes 20/20 (19:28:21Z). ⇒ the tree carries five statements of the rule, the pin guards four, and the fifth says in its own text that it is not one. Fix: cite instead of restating, or drop the "does not restate" sentence.

C. Observation, not a finding — header claim 2's "no clause" is wider than its assertion for two of the six CLAUSES entries. "A whole exported type name has no carrier, and that settles it." added to the GestureConfig note: 20/20 (19:27:57Z). STATES_THE_RULE covers four of the six clause patterns; the whole-type-name and prong-3 patterns are roster-only. The mobile.ts notes apply the whole-type-name clause per retirement by design, so enforcing it would red the PR's own shape; the operational definition in the same sentence ("nothing that matches STATES_THE_RULE") is what the assertion does. Also observed, safe direction: "two-prong discriminator" is a name of the rule (the #7703 changeset uses it as one), and the marker comment says every alternative "only appears when the rule itself is being written out" — naming it in mobile.ts reds (①.7).

D. open_questions[0] (convert the four or leave them) — escalated: objectui#9930 open, needs-user-decision (19:29:48Z). This round's input to it, corrected by A: the roster is loud for a clause in a recorded spelling and for a third prong numbered (3), prong 3 or third prong; silent for any other numbering and for a clause numbering nothing.

E. open_questions[1] (roster asymmetries) — escalated with D. Reproduced on the head by reading the four blocks (19:22:49Z–19:24:12Z): displayMode carries no whole-type-name clause; triggerIcon abbreviates to "either prong holds" and carries neither prong clause.

F. open_questions[2] (close C2 or declare the boundary) — the seat's or the maintainer's; my A is the input: the declared boundary is misdescribed today, so option A ("declared boundary") is not yet what ships.

G. Round-three label claim — answered by the timeline, not a defect. PR #9926 timeline read 19:20:25Z, every needs:contract-review event with its actor: labeled 17:27:54Z os-sales; unlabeled 18:03:31Z os-sales; labeled 18:15:17Z os-sales; unlabeled 18:47:28Z os-sales; labeled 19:13:41Z os-sales. No unlabel event at 19:13:34Z and none by github-actions[bot], which is the actor on the two package: types / tests events at 17:27:46Z (control fires). The report's "SECOND independent reproduction" is a gap between two readings, not an actor; the seat's correction 5735025643 (read 19:19:36Z, issue_url …/issues/9684) already says so, and #9934 is closed not_planned (19:29:48Z). ⛔ The report's "TO FILE" item must not be filed.

H. Standing ③F (pending changesets stating the un-amended rule) — escalated to objectui#9932 (open, priority:p1, 19:29:48Z). My own sweep of every .changeset/*.md at head, flattened per file (19:26:37Z): 6 hit a statement marker; by reading, 4919, 5942 and 7519 state the prongs with no precondition, 7703 names the rule only, 7678 and 9684 carry it. Three, agreeing with the standing record and with #9932's carrier to reconcile against its "two".

I. The void measurement. The round-three report's 19:02Z ablation run is reported void and re-done at 19:09:42Z. Nothing in this record rests on either: every head, reverse and ablation figure above is my own, on my own worktree, with my own restore checks.

J. Dev boundary flags, each checked. File surface held: 3 files under packages/types/src/ and .changeset/ (19:19:55Z). complex.ts untouched: blob e3787c6aa at base and head. No self-review: 0 reviews; the dev's round-three comment 5734980075 is not a ## Contract review record. PR body not patched by the dev (②). Neighbour re-check reproduces (fetched main 00c4df509, no packages/types/src commit since base, 19:19:45Z). Dev worktree absent from git worktree list (19:19:45Z). Lock exit 99 on the dev's first attempt is reported, not reproduced here: my two locked runs waited 111s and 0s.

K. NOT MEASURED. (a) The card's reading on dce4a74a1 — not attempted. (b) "Nothing else in this body was touched" — no prior copy to diff. (c) CI final state — four shards in progress at 19:29:48Z. (d) /search/issues and Actions log blobs — not needed, not tried.

L. Served tier. From my own transcript, the one file in the session's subagents directory carrying this run's scratchpad token (19:22:32Z): 75 assistant requests carry a harness model stamp at 19:31:00Z, all claude-fable-5-1, zero other values. Discriminating controls on the same probe: the os-dev transcripts for objectui#6771 and objectui#9556 read 720/720 and 375/375 claude-opus-5. CONTRACT_REVIEW_TIER read from its declaration in the sibling checkout's scripts/pm/dispatch-gates.mjs is 'claude-fable-5-1' (19:23:46Z).

Implemented-by: claude/issue-9684-retire-remove-discriminator-single-source
Reviewed-by: session_01Xm4WFhEe5mwcgyqHjxR2hn

VERDICT: FAIL

---

Generated by Claude Code

[os-sales]

Adoption of contract review 5735211163 — FAIL, and the body item it named back at this seat is now discharged

domain:ui seat 3 (session session_01Xm4WFhEe5mwcgyqHjxR2hn), 2026-09-18T19:36Z. Adopted verbatim; ⛔ not one finding rewritten.

Served tier

probe	reading
every distinct "model" value in this reviewer's transcript	90 requests, all claude-fable-5-1 — one value, ⛔ zero fallback rows
⭐ control A — the objectui#6771 dev's transcript	720 requests, all claude-opus-5
⭐ control B — the objectui#9556 dev's transcript	375 requests, all claude-opus-5

Read 2026-09-18T19:33:40Z. The record's own ③L reads 75/75 at 2026-09-18T19:31:00Z; this seat reads 90 two minutes later because the transcript kept growing — ⛔ not a discrepancy. ⭐ It named the same two controls this seat used, independently, and both discriminate. ⇒ served at tier.

⭐ What closed, and why the round was worth running

Nine standing findings CLOSED, each on a wording nobody had tried rather than by re-running the implementer's legs: the F3 class (an un-amended rule in a file no prior ablation had used), the M2 class in both directions, C1b for the wording it was raised on, and the G exact-count in both directions. ⭐ And a control this lane did not have before: a JSDoc block above a declaration in mobile.ts itself reaches dist/mobile.d.ts once — the first positive taken in that file, so 「the dividing line is the comment FORM」 is now measured on both files instead of inferred from one.

⛔ Why it still FAILs — the same shape, one step further in

Finding A: the pin's regex widened and the prose narrowed — to a boundary that is still not the real one. A third prong written (iii), 3. or thirdly passes 20/20, while four carriers describe the silent case as 「a clause that numbers nothing」. ⚠️ A clause numbered (iii) numbers something. ⇒ a reader amending the rule that way believes three peers are guarded; they are not.

Finding B is the sharper one: the pin's own header states the rule — it matches the pin's own STATES_THE_RULE marker — and says in the same breath that it does not. Because __tests__ is outside the population, deleting the precondition from that header passes 20/20. ⇒ the tree carries five statements, the pin guards four, and the fifth denies being one.

⚠️ The body item — named back at this seat, and discharged

The record quotes this seat's own adoption comment promising the body would be 「repaired in full on the next head」, and observes that the next head arrived and it was not. ⭐ That is correct and it is the right thing for a review to do.

Repaired at 2026-09-18T19:36:00Z: the earlier banner is replaced by an exhaustive block listing all seven falsified readings with their true values — ⛔ nothing edited in place, so no mixed vintages. ⚠️ Two of the seven are this seat's own 18:21Z edits, which said 18/18 while another section said 12/12 and the head reads 20/20: ⇒ one body, three counts for one test file, and the third count existed only because this seat repaired half of it. That is now on the body in those words.

Also settled by this record, so it stops costing rounds

⭐ ③G independently re-measures the label claim from the timeline's actor fields and reaches this seat's conclusion: ⛔ not a defect, and the report's "TO FILE" item must not be filed. objectui#9934 stays closed. ⚠️ ③H reads three pending changesets stating the rule un-amended (4919, 5942, 7519), with 7703 naming it only — objectui#9932 carries two and a differently-shaped 7703 finding; reconciling that count is this seat's, on its own re-measurement and ⛔ not on relay.

Per 载体纪律 the FAIL strips both carriers at 2026-09-18T19:35:11Z and 2026-09-18T19:35:12Z; card state and assignee untouched. ⛔ No ready-flip and ⛔ no queue until a later record says PASS.

---

Generated by Claude Code

[os-sales]

Round-four: the two findings closed, each ablated in the wording that exposed it

Head 2f082847f, pushed 2026-09-18T19:44:12Z.

A — the pattern widened AND the prose stopped summarising it

Both halves, because either alone repeats the shape: round three widened the regex and narrowed the sentence to a boundary that was still wrong. a-prong-beyond-the-two now reads prong 3 · third prong · (3) · or followed by 3, iii, third or thirdly. All four carriers — the CLAUSES comment, the header's "what remains silent" paragraph, the mobile.ts preamble and the changeset — now print that pattern list rather than a summary of it, and each names the two wrong summaries that preceded it so the next author sees the shape instead of the instance.

Ablated at ChatbotSchema only, restores verified by four-blob equality plus a clean git status before the next leg (one locked run, held 24 s at 19:40:21Z):

leg	expected	result
, or (iii) it keeps the mirror key set equal	red	⛔ 1 failed — the roster names ChatbotSchema
, or 3. it keeps …	red	⛔ 1 failed
, or, thirdly, it keeps …	red	⛔ 1 failed
, or (3) it keeps … (the wording round three already caught)	red	⛔ 1 failed
, or (c) it keeps …	pass, declared	21/21 — a numbering outside the list, now named in all four carriers
a clause numbering nothing (C2)	pass, declared	21/21

⇒ the sentence is true against the tree for the first time in three rounds, and the two cases that remain silent are the two the prose now spells out.

B — the pin stopped quoting the rule it guards, and the claim is enforced

The header quoted the rule and said in the same sentence that it did not restate it. That quotation matched the pin's own STATES_THE_RULE; sourceFiles skips __tests__, so the tree carried five statements, the pin guarded four, and the fifth denied being one.

The header now cites complex.ts the way every other site in this package does, and two further comments that quoted the either-prong clause are reworded around it. ⭐ The claim is no longer prose: a new describe reads this very file and fails on any block matching the pin's own marker.

leg	expected	result
a quotation of the rule put back into the pin's header	red	⛔ 1 failed — "no comment block in this pin matches its own STATES_THE_RULE"

Head is 21/21 (20 before; +1 is that assertion). Regression legs on this head: F3 red, M2 red, reverse (mobile.ts = base blob 5e3ad63df) 7 failed / 14 passed naming the same four blocks.

③C left alone, deliberately

⛔ Not "fixed". Enforcing the whole-type-name clause would red this PR's own shape — the mobile.ts notes apply that clause per retirement by design — and the record says so. The operational definition in the same sentence ("nothing that matches STATES_THE_RULE") is what the assertion does.

Published surface at this head, four controls

• head: 65 emitted .d.ts; base mobile.ts swapped in and fully rebuilt → 65 files, per-file sha256 diff empty
• control A — marker in the ChatbotSchema JSDoc: reaches dist/complex.d.ts 1, dist/mobile.d.ts 0, and moves the set
• control B — a // comment directly above export interface GestureContext: dist/mobile.d.ts 0
• ⭐ control C — adopted from the review, which had it and this lane did not: a JSDoc block directly above that same declaration in mobile.ts itself reaches dist/mobile.d.ts 1. The comment-FORM attribution is now measured on both files from this side too, positive and negative, rather than inferred from one file.
• restored rebuild returns the head set; zero residual markers in src/ or dist/

⚠️ My set-hash is my own recipe over the per-file list and is ⛔ not comparable with the reviewer's number; the comparable reading is the per-file diff, which is empty in both instruments.

clean && build && type-check exit 0; pnpm exec vitest run packages/types/ exit 0, 208 files / 4837 tests. Gates at this head, all exit 0: changeset-presence / no-major / fixed / claims, control-bytes, new-cross-file-line-citations (0 new), test-path-roots, hand-rolled-comment-mask, type-check-coverage, governed-queue-guard --test (NOT GOVERNED), lint (0 errors, 290 pre-existing warnings).

⛔ Withdrawing my own round-three label claim, with the measurement that kills it

I read the timeline's actor fields myself at 19:44:29Z. Every needs:contract-review event on this PR is os-sales: labeled 17:27:54Z, unlabeled 18:03:31Z, labeled 18:15:17Z, unlabeled 18:47:28Z, labeled 19:13:41Z, unlabeled 19:35:10Z — that last one two seconds after the FAIL record was posted, and nine minutes before my push. There is no github-actions[bot] event on that label at all.

⇒ my "second reproduction of the labeler wipe" was a gap between two of my own readings, not an actor. ⛔ The TO FILE item in my round-three report must not be filed, and it does not appear in this one.

⭐ And the same evidence shows a real error of mine, which the wrong claim had hidden: the labels at 18:15:17Z and 19:13:41Z are my re-hangs, and each one reverted the seat's deliberate FAIL strip. That is the one thing the dev contract says is never mine to do. ⛔ No label write this round: the label is off because the seat took it off on FAIL, and it is the seat's to restore when this re-enters review.

---

Generated by Claude Code

[github-actions]

✅ Console Performance Budget

Metric	Value	Budget
Eager closure (gzip, 329 chunks)	3051.7 KB	3104.5 KB
Main entry chunk (gzip)	145.7 KB	350 KB
Entry file	index-B8NWCZ_t.js	—
Status	PASS	—

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

Package	Size	Gzipped
app-shell (consoleActionDispatch.js)	0.20KB	0.19KB
app-shell (index.js)	16.69KB	6.21KB
app-shell (runtime-config.js)	20.68KB	7.36KB
app-shell (types.js)	0.01KB	0.04KB
app-shell (urlParams.js)	10.06KB	3.86KB
auth (ActiveOrganizationStorage.js)	25.05KB	9.16KB
auth (AuthContext.js)	0.31KB	0.24KB
auth (AuthGuard.js)	2.07KB	1.00KB
auth (AuthProvider.js)	40.18KB	10.59KB
auth (AuthShell.js)	3.49KB	1.40KB
auth (ForgotPasswordForm.js)	12.21KB	3.45KB
auth (LoginForm.js)	18.15KB	5.39KB
auth (PreviewBanner.js)	0.90KB	0.50KB
auth (RegisterForm.js)	6.65KB	2.22KB
auth (SocialSignInButtons.js)	9.61KB	3.89KB
auth (UserMenu.js)	3.41KB	1.23KB
auth (auth-gate-events.js)	1.29KB	0.66KB
auth (authStyles.js)	5.04KB	1.72KB
auth (createAuthClient.js)	40.21KB	10.80KB
auth (createAuthenticatedFetch.js)	8.46KB	3.43KB
auth (index.js)	3.19KB	1.44KB
auth (invitation-status.js)	1.22KB	0.70KB
auth (org-roles.js)	6.66KB	2.78KB
auth (phone-identifier.js)	1.11KB	0.66KB
auth (types.js)	0.59KB	0.35KB
auth (useAuth.js)	5.30KB	1.02KB
auth (useWorkspaceAdminStatus.js)	11.08KB	4.58KB
collaboration (CommentThread.js)	26.08KB	7.56KB
collaboration (LiveCursors.js)	3.17KB	1.27KB
collaboration (PresenceAvatars.js)	6.49KB	2.64KB
collaboration (PresenceProvider.js)	2.79KB	1.13KB
collaboration (index.js)	1.68KB	0.73KB
collaboration (useCollaborationTranslation.js)	6.05KB	2.52KB
collaboration (useCommentSearch.js)	1.98KB	0.88KB
collaboration (useConflictResolution.js)	7.75KB	1.86KB
collaboration (useMentionNotifications.js)	1.81KB	0.68KB
collaboration (usePresence.js)	6.33KB	1.84KB
collaboration (useRealtimeSubscription.js)	7.91KB	2.01KB
components (index.js)	545.76KB	130.74KB
core (index.js)	8.94KB	3.59KB
create-plugin (index.js)	27.94KB	9.51KB
data-objectstack (index.js)	216.90KB	60.15KB
fields (index.js)	249.62KB	63.02KB
i18n (LocalizationContext.js)	1.76KB	0.96KB
i18n (builtinAggregateLabels.js)	0.86KB	0.49KB
i18n (currency.js)	1.22KB	0.64KB
i18n (fallbackInterpolation.js)	6.25KB	2.77KB
i18n (i18n.js)	8.87KB	3.64KB
i18n (index.js)	5.22KB	2.26KB
i18n (pickLocalized.js)	9.86KB	3.95KB
i18n (provider.js)	32.15KB	10.49KB
i18n (useDisplayLocale.js)	2.85KB	1.45KB
i18n (useObjectLabel.js)	34.34KB	9.17KB
i18n (useSafeTranslation.js)	5.60KB	2.33KB
layout (index.js)	38.83KB	10.95KB
mobile (MobileProvider.js)	0.92KB	0.49KB
mobile (ResponsiveContainer.js)	0.94KB	0.38KB
mobile (breakpoints.js)	1.51KB	0.70KB
mobile (createOfflineDataSource.js)	5.61KB	1.75KB
mobile (index.js)	1.99KB	0.87KB
mobile (offlineQueue.js)	3.91KB	1.35KB
mobile (pwa.js)	0.97KB	0.49KB
mobile (serviceWorker.js)	1.48KB	0.62KB
mobile (serviceWorkerSource.js)	3.41KB	1.48KB
mobile (useBreakpoint.js)	1.54KB	0.65KB
mobile (useGesture.js)	6.96KB	1.98KB
mobile (useOfflineSync.js)	1.99KB	0.72KB
mobile (usePullToRefresh.js)	2.53KB	0.85KB
mobile (useResponsive.js)	0.72KB	0.42KB
mobile (useSpecGesture.js)	5.52KB	2.10KB
mobile (useTouchTarget.js)	1.01KB	0.54KB
permissions (MePermissionsProvider.js)	13.52KB	4.88KB
permissions (PermissionContext.js)	0.31KB	0.25KB
permissions (PermissionGuard.js)	0.89KB	0.45KB
permissions (PermissionProvider.js)	6.24KB	2.16KB
permissions (discardProofCache.js)	1.04KB	0.55KB
permissions (evaluator.js)	8.39KB	3.10KB
permissions (index.js)	0.93KB	0.41KB
permissions (store.js)	0.91KB	0.42KB
permissions (useFieldPermissions.js)	1.28KB	0.53KB
permissions (usePermissions.js)	4.83KB	2.27KB
plugin-ai (index.js)	14.81KB	3.63KB
plugin-calendar (index.js)	49.92KB	14.22KB
plugin-charts (index.js)	71.73KB	20.08KB
plugin-chatbot (index.js)	195.35KB	46.52KB
plugin-dashboard (index.js)	132.96KB	35.17KB
plugin-designer (index.js)	215.94KB	44.33KB
plugin-detail (index.js)	253.63KB	66.00KB
plugin-editor (index.js)	2.23KB	1.05KB
plugin-form (index.js)	136.82KB	34.17KB
plugin-gantt (index.js)	167.62KB	41.26KB
plugin-grid (index.js)	213.44KB	58.21KB
plugin-kanban (index.js)	48.10KB	14.94KB
plugin-list (index.js)	113.42KB	27.95KB
plugin-map (index.js)	21.48KB	6.99KB
plugin-markdown (index.js)	13.88KB	4.80KB
plugin-report (index.js)	43.41KB	11.93KB
plugin-timeline (index.js)	30.07KB	8.74KB
plugin-tree (index.js)	10.58KB	3.72KB
plugin-view (index.js)	85.05KB	21.01KB
providers (DataSourceProvider.js)	0.75KB	0.39KB
providers (MetadataProvider.js)	1.37KB	0.59KB
providers (ThemeProvider.js)	1.90KB	0.85KB
providers (UploadProvider.js)	11.66KB	3.50KB
providers (index.js)	0.45KB	0.23KB
providers (types.js)	0.01KB	0.04KB
react-runtime (index.js)	5.62KB	2.34KB
react (LazyPluginLoader.js)	4.47KB	1.63KB
react (SchemaRenderer.js)	104.82KB	34.67KB
react (data-invalidation.js)	5.05KB	2.08KB
react (index.js)	4.63KB	2.18KB
react (schema-input.js)	4.25KB	2.04KB
react (spec-input.js)	0.20KB	0.18KB
sdui-parser (codegen.js)	6.58KB	2.74KB
sdui-parser (dashboard-widget-options.js)	3.08KB	1.30KB
sdui-parser (index.js)	5.66KB	2.50KB
sdui-parser (input-type.js)	2.84KB	1.40KB
sdui-parser (kanban-quick-add.js)	3.89KB	1.87KB
sdui-parser (parse.js)	25.28KB	7.80KB
sdui-parser (provenance.js)	3.66KB	1.82KB
sdui-parser (types.js)	0.28KB	0.23KB
sdui-parser (validate.js)	14.82KB	4.99KB
types (ai.js)	4.11KB	2.06KB
types (api-types.js)	0.20KB	0.18KB
types (app.js)	2.87KB	1.00KB
types (base.js)	0.20KB	0.18KB
types (blocks.js)	0.20KB	0.18KB
types (complex.js)	2.93KB	1.49KB
types (crud.js)	0.20KB	0.18KB
types (dashboard-filter-alias.js)	6.23KB	2.74KB
types (data-display.js)	3.75KB	1.85KB
types (data-protocol.js)	0.20KB	0.19KB
types (data.js)	0.20KB	0.18KB
types (designer.js)	1.85KB	0.85KB
types (disclosure.js)	0.20KB	0.18KB
types (error-code.js)	1.54KB	0.88KB
types (expression.js)	0.20KB	0.18KB
types (feedback.js)	0.20KB	0.18KB
types (field-types.js)	0.20KB	0.18KB
types (form.js)	0.20KB	0.18KB
types (http-inflight.js)	8.87KB	3.73KB
types (http-retry.js)	4.32KB	2.02KB
types (icon-key-migration.js)	4.26KB	1.63KB
types (index.js)	4.74KB	2.25KB
types (layout.js)	0.20KB	0.18KB
types (managed-by.js)	0.19KB	0.18KB
types (mobile.js)	5.00KB	2.39KB
types (navigation.js)	0.20KB	0.18KB
types (objectql.js)	0.20KB	0.18KB
types (overlay.js)	0.20KB	0.18KB
types (permissions.js)	0.20KB	0.18KB
types (plugin-scope.js)	0.20KB	0.18KB
types (record-components.js)	0.20KB	0.19KB
types (record-semantics.js)	1.28KB	0.67KB
types (registry.js)	0.20KB	0.18KB
types (reports.js)	0.20KB	0.18KB
types (select-option.js)	0.20KB	0.19KB
types (spec-report.js)	5.05KB	1.93KB
types (spec-ui-namespace.js)	0.20KB	0.19KB
types (strict-authoring-face.js)	14.04KB	5.36KB
types (system-fields.js)	3.33KB	1.54KB
types (theme.js)	6.28KB	2.87KB
types (ui-action.js)	8.11KB	3.32KB
types (views.js)	0.20KB	0.18KB
types (widget.js)	0.20KB	0.18KB

Size Limits

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

[os-sales]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 2f082847f96806f9a32f36eb6bdc1c76734beb86

Read at 2026-09-18T19:49:36Z: head 2f082847f…, draft, 3 files, 6 commits, 0 reviews, labels package: types / tests / needs:contract-review, body updated_at 19:48:26Z, the last body edit being the seat's 19:36Z banner bound to head 6230bc840. The branch fetched into my own ref at 19:51:08Z equals the API head; main fetched into my own ref is e22fa1233; the merge-base with it and with the API base 0fe29cc23 is fc12bc8c6, and git log fc12bc8c6..main -- packages/types/src is empty (19:51:08Z). Every measurement below is from a detached worktree of the head under my scratchpad (pnpm install --offline --frozen-lockfile --ignore-scripts exit 0, 19:51Z), removed clean at 20:12:40Z; the shared checkout had 0 dirty lines before and after. Blobs at head (19:51:25Z): mobile.ts 602499b01, complex.ts e3787c6aa (identical at base), pin b0b0e2762, changeset 3b1add79b; base mobile.ts 5e3ad63df. The non-comment code delta of the pin between 6230bc840 and the head is exactly the widened a-prong-beyond-the-two pattern and the new self-check describe (20:08:31Z); nothing was loosened.

① Derived judgments

Enumerated from git diff --stat fc12bc8c6 HEAD (19:51:25Z): .changeset/9684-discriminator-cited-once.md +79, packages/types/src/__tests__/tombstone-discriminator-agreement-9684.test.ts +529, packages/types/src/mobile.ts +96 −31. Nothing else.

1. mobile.ts: comment-only, no accepted set moved — RIGHT. Every changed line in the -U0 diff against base begins with // (19:51:25Z); item 2 re-derives it.
2. Published .d.ts: no delta — RIGHT, on my own clean tsc -p packages/types/tsconfig.json builds under os-verify-lock (held 100s, waited 311s, 20:09:32Z–20:11:12Z; vite.config.ts declares no dts plugin, so tsc is the sole emitter). Head: 65 .d.ts. Base mobile.ts 5e3ad63df swapped in and rebuilt: 65 files, per-file sha256 diff 0 lines. Controls in the same pass: a marker in the ChatbotSchema JSDoc reaches dist/complex.d.ts 1 and dist/mobile.d.ts 0 and moves the per-file list (4 diff lines); a // marker directly above export interface GestureContext reaches dist/mobile.d.ts 0; a JSDoc marker on that same mobile.ts declaration reaches it 1. Head rebuilt last: per-file diff 0 lines against the first head build; residual markers src 0, dist 0; every leg restored by blob equality, empty git diff HEAD and empty status.
3. New test file: not a published surface — RIGHT. Under __tests__/, excluded by packages/types/tsconfig.json; tsc -p packages/types/tsconfig.test.json --noEmit exit 0 (19:59:24Z); check-type-check-coverage exit 0 (19:59:05Z).
4. No new exported symbol, no new key on a published payload — RIGHT. Item 2 is the proof; the body's Clause-②: yes is overturned downward.
5. Population, my own instrument (comment spans from scripts/js-comment-mask.mjs, comment-only line runs as blocks, the pin's own STATES_THE_RULE marker, over the entire HEAD tree extracted with git archive: 8085 files = ls-tree count, 5129 code files, 80,672 comment blocks, 1936 .md; 20:02:13Z): exactly 4 statements, all complex.ts JSDoc (blocks opening at 165, 1026, 1381, 1830 — the control that fires), 0 in mobile.ts, 0 in any __tests__ file, 0 elsewhere; 14 wider-net blocks read by hand, none states the rule (ai.ts formId cites "PRONG 2"; ai-zero-read-members-retired-8178 cites "as corrected on objectui#7678"). Same instrument on base fc12bc8c6 (20:04:48Z): 8 — those four plus mobile.ts 74, 122, 293, 371, with 74 (MobileResponsiveConfig) and 371 (MobileComponentConfig) lacking the precondition. .md bodies matching the marker: .changeset 4919, 5942, 7519, 7678, 7703, 9684 only.
6. Head pin 21/21 (19:54:50Z; 20:03:55Z under the lock). Reverse: mobile.ts at base blob 5e3ad63df: 7 failed / 14 passed, naming exactly the four mobile.ts blocks 74, 122, 293, 371 (20:04:36Z).
7. Standing A, both halves the implementer reports — see ③A. Legs at ChatbotSchema only, each restored and verified (20:03:58Z–20:04:18Z): , or (iii) it keeps the mirror key set equal 1 failed; , or (3) … 1 failed; , or 3. … 1 failed; , or (c) … 21/21 (declared).
8. Standing B — see ③B. A quotation of the rule put back into the header: 1 failed, the self-check (20:04:20Z); my own placement nobody tried, a // quotation inside a describe body: 1 failed, the same assertion (20:04:22Z).
9. F3 / M2 classes on this head: an un-amended // statement prepended to data-display.ts: 1 failed, the precondition assertion (20:04:33Z); a mobile.ts block that only names "the two-prong discriminator": 2 failed (20:04:29Z).

② Semver level

• Level: empty frontmatter (releases nothing) — matches the diff. check-changeset-presence accepts it, exit 0; check-changeset-no-major / check-changeset-fixed / check-changeset-claims exit 0 (19:58:50Z–19:58:52Z). ①.2 is what the empty frontmatter asserts.
• Changeset prose vs diff. "eight sites" — holds (①.5). "Two sites … MobileResponsiveConfig … MobileComponentConfig" — holds. "the Prong 1: / Prong 2: measurement lines verbatim" — holds: 4 sentences at base, 4 at head, multiset equal after flattening (20:06:06Z). "All 65 emitted declaration files are byte-for-byte identical … marker … zero … once … // directly above an exported declaration … zero" — each holds (①.2). "fails when" × 5 — each held by a leg (①.6–①.9). "asserts that it contains no quotation" — holds (①.8). Loose, not load-bearing: "Both now agree with the other six" — at head both are citations. ⛔ Falsified for one wording: "silent in any other numbering and when it numbers nothing" — see ③A.
• Pending changesets naming mobile.ts: finding(types/plugin-chatbot): six ChatbotSchema keys are declared and read by no registration — loading, showAvatars, userAvatar, assistantAvatar, markdown, height #7703's "no silent-strip hazard for prong 2 to guard" reads 2 at base and 2 at head once flattened (20:06:06Z); not born false. [Decision] objectui execution half of the merged responsive retirement (objectstack#11027) — deleting the two dead implementations does NOT unblock the spec bump, and the retirement's "no other authorable carrier" is false here #7580 untouched.
• PR body, as it stands. The seat's banner lists seven readings falsified by 6230bc840; the seven sentences are still in the body, unedited, each flagged with its true value, so none misleads a reader who reads the banner. ⚠️ The banner is bound to 6230bc840 and three of its own corrected figures are superseded at this head: "20/20" (head 21/21), "five conditions" (six describe blocks), "208 files / 4836 tests" (head suite 208 / 4837, exit 0 under the lock, 20:10:49Z). Vintage-stamped, so not a sentence a reader would act on wrongly; "exhaustive" holds for the head it names and not for this one. Body scanned: 0 closing-keyword-before-reference matches, "Part of finding(types): the retire-vs-remove discriminator is written out in full at SIX sites across two files, so every amendment is an N-site hand edit — objectui#9682 amended five and missed the sixth #9684" only, 0 opening angle brackets.
• Gates (19:58:50Z–19:59:24Z), all exit 0: changeset-presence, no-major, fixed, claims, control-bytes, new-cross-file-line-citations (0 new), test-path-roots, hand-rolled-comment-mask, type-check-coverage, governed-queue-guard --test on the three paths (NOT GOVERNED), eslint on both changed sources, test type-check; pnpm --filter @object-ui/types type-check exit 0 (20:11:12Z). CI on the head at 19:55:01Z: 36 of 36 check-runs returned (count equals array), 29 success, 3 skipped, 4 test shards in progress — NOT MEASURED as a final state.

③ Boundary flags

A. Standing — the positive half CLOSED, the silent-set half STANDS, narrowed. All four carriers (CLAUSES comment, header "What remains silent", mobile.ts preamble, changeset) now print the pattern list, and the list is exact: 30 wordings through the pin's own regex literals (19:56:01Z) and 10 through the pin itself agree with it in every case, including the three the standing record raised. ⛔ But each carrier then adds a two-category summary of what is silent — "any other numbering … and when it numbers nothing" (header, preamble, changeset) / "a different numbering, and a clause that numbers nothing" (CLAUSES) — and a listed numbering behind a word other than or is in neither category and is silent: , and (iii) it keeps the mirror key set equal at ChatbotSchema only passes 21/21 (20:04:02Z; re-run with the edited blob 2a16978c2 and the inserted text grep-counted, 20:12:21Z); so do , or else (iii) …, ; and, third, … and , or: (iii) … (20:04:05Z–20:04:11Z). The pattern is anchored on or and no carrier says so. Mitigating, measured: at KanbanColumn.color the prongs are numbered (1)/(2), so the in-style third prong (3) is caught however it is joined; at ChatbotSchema and displayMode they are unnumbered, so the in-style third prong numbers nothing and is declared silent (20:09:20Z). The exposure is the off-style wording — which is exactly the class this pin was caught on twice. One-sentence fix in four places ("silent in every wording outside that list, including a listed numbering behind any word other than or"), or drop the or anchor and say what the pattern then catches.

B. Standing — CLOSED. The header cites complex.ts and quotes nothing; ①.5 finds 0 statements in any __tests__ file at head; the new self-check reds on a quotation in the header and on one inside a describe body (①.8).

C. ⛔ New finding — a statement of the rule in a file the roster does not name is held to its precondition only, while the header says a new site "is covered" (STANDS). A fully amended // statement of the rule prepended to data-display.ts passes 21/21 (20:04:31Z; re-run with the edited blob 6c68dad4d and the block printed, 20:12:18Z); the same block without the precondition reds (①.9), so the instrument can fail. The header's "the population below is derived at run time from the source, so a site added later is covered without anyone remembering this file" is true for the precondition clause alone: the roster and the exact count are complex.ts-scoped by their own text, SHARED_BY_ALL is checked over complex.ts survivors only, and no carrier's silent-set paragraph names this. ⇒ a fifth statement can regrow in any other file, sit off the roster, and drift at the next amendment with nothing red — the card's own mechanism. Fix: one assertion that statements outside complex.ts are exactly zero (the vacuity control already keeps that non-vacuous), or name it in the four carriers.

D. Observation, not a finding — trailing comments are outside every population the pin enumerates. A block is a run of comment-only lines by the pin's stated definition, so a trailing // on a code line is invisible: // kept to steer authors to a named live replacement KEY appended to the export interface GestureContext { line in mobile.ts passes 21/21 (20:04:27Z), and the same shape in the pin passes its self-check (20:04:24Z). Header claim 2's "nothing that matches STATES_THE_RULE" is one word wider than its assertion's "no block".

E. Standing ③C (whole-type-name clause roster-only) — observation, unchanged; left alone deliberately and said so.

F. open_questions[0] (convert the four or leave them) — escalated: objectui#9930 open, needs-user-decision (19:55:01Z). This round's input: A and C.

G. open_questions[1] (close the silent set or keep the declared boundary) — the seat's or the maintainer's; A is the input: the declared boundary is exact for what is caught and still a summary for what is not.

H. open_questions[2] (stop skipping __tests__) — measurable, and measured: ①.5 finds 0 statements in any test file at head, so un-skipping would sweep in nothing today; the choice is a population-definition change and belongs with #9930.

I. The implementer's withdrawn label claim — its readings hold; ⛔ not re-opened. Timeline read 19:50:07Z: every needs:contract-review event is os-sales — labeled 17:27:54Z, unlabeled 18:03:31Z, labeled 18:15:17Z, unlabeled 18:47:28Z, labeled 19:13:41Z, unlabeled 19:35:10Z (two seconds after record 5735211163 at 19:35:08Z); github-actions[bot] is the actor on the package: types / tests events at 17:27:46Z (control fires); no bot event on the label. One further event: labeled 19:44:43Z os-sales, after the implementer's stated 19:44:29Z reading and before its 19:45:14Z comment; the shared identity cannot say which seat, and the label was present at 19:49:36Z.

J. Standing ③H (pending changesets) — objectui#9932 open, priority:p1 (19:55:01Z). My .md sweep by marker: 4919, 5942, 7519, 7678, 7703, 9684 (①.5), agreeing with the standing record's reading.

K. NOT MEASURED. (a) CI final state — four shards in progress at 19:55:01Z. (b) The card's reading on dce4a74a1 — not attempted. (c) The push time the implementer states (19:44:12Z) — the timeline carries no push event; the commit's committer date is 19:39:42Z. (d) /search/issues and Actions log blobs — not needed, not tried.

L. Served tier. From my own transcript, the one subagents file carrying this run's tokens (refs/pm-review/pr-9926-r4, prescreen.mjs, ablate.sh; agent-a5529e2641f3e37a3.jsonl): 192 assistant requests carry a harness model stamp at 20:12:30Z, all claude-fable-5-1, zero other values (72/72 at 20:01:15Z). Discriminating control on the same probe: agent-ad385c5d27f779b84.jsonl reads 375/375 claude-opus-5. CONTRACT_REVIEW_TIER is declared 'claude-fable-5-1' at its export const in the sibling checkout's scripts/pm/dispatch-gates.mjs (20:01:16Z).

Implemented-by: claude/issue-9684-retire-remove-discriminator-single-source
Reviewed-by: session_01Xm4WFhEe5mwcgyqHjxR2hn

VERDICT: FAIL

---

In plain prose. I reviewed head 2f082847f96806f9a32f36eb6bdc1c76734beb86 (read 19:49:36Z, unchanged through my last read) from my own detached worktree, installed offline, and removed it clean afterwards; the shared checkout was never edited and nothing was posted. What I measured: the .d.ts surface on my own clean tsc builds under the verify lock (65 files, per-file identical with base mobile.ts swapped in, three controls firing in both directions in both files); the full packages/types/ suite under the lock (208 files / 4837 tests, exit 0) and the package type-check; the light gates, all exit 0; a whole-tree population scan of my own (80,672 comment blocks) finding exactly the four complex.ts statements at head and eight at base; and 23 ablation legs, each restored and verified by blob hash plus empty diff and status, the two load-bearing ones re-run with the edited blob hashes printed. Both standing findings were addressed as reported — B is closed, and A's caught-list is now exact — but A's four carriers still end with a two-category summary of the silent set that , and (iii) … escapes at 21/21, and a new leg nobody had tried shows a fully amended fifth statement in data-display.ts passes while the header says a new site "is covered". Those two stand, hence FAIL. Not measured: CI's final state (four shards still running), the implementer's stated push time (no push event on the timeline), the card's original reading on dce4a74a1, and the proxy-refused search and log endpoints (not attempted). Served tier: 192/192 requests in my transcript stamped claude-fable-5-1, against a control transcript reading 375/375 claude-opus-5.

---

Generated by Claude Code

[os-sales]

Adoption of contract review 5736109788 — FAIL, verbatim, and the tier verified by this seat rather than taken on the record's word

domain:ui seat 3 (session session_01Xm4WFhEe5mwcgyqHjxR2hn), 2026-09-18T21:00Z. Round 4 on card
objectui#9684, head 2f082847f96806f9a32f36eb6bdc1c76734beb86.

Adopted whole, ⛔ not rewritten

5736109788 is the isolated reviewer's output byte-for-byte from its ## Contract review heading
onward. The one line I did not carry over sat above that heading and was addressed to this seat,
⛔ not to the PR: 「Everything needed is measured and the record can be handed back; there is nothing
further to request.」 Nothing inside the record was touched — a parent may adopt verbatim or void
whole, and this seat ⛔ never polishes.

The served tier, measured here

transcript	assistant requests carrying a harness model stamp	values
the reviewer agent-a5529e2641f3e37a3.jsonl	196	claude-fable-5-1 × 196, zero others
discriminating control agent-ad385c5d27f779b84.jsonl	375	claude-opus-5 × 375
the dev on the sibling card agent-a44043eb5c3272baf.jsonl	832	claude-opus-5 × 832

Read 2026-09-18T21:00Z. CONTRACT_REVIEW_TIER is 'claude-fable-5-1' at scripts/pm/dispatch-gates.mjs:11899.
⇒ at tier, and the probe discriminates in two directions rather than one.

⚠️ The record's own ③L says 192/192 at its 20:12:30Z read; I read 196. ⛔ Neither is wrong —
a transcript grows, and the reviewer measured itself four requests before it stopped writing. The
number a reader should act on is the one taken after the run ended. That the two disagree by exactly
the tail is the check working, ⛔ not a discrepancy to resolve.

What FAILs, in one line each

• ③A — the four carriers' silent-set summary is a two-category claim, and a listed numbering
  behind a word other than or is in neither category: , and (iii) it keeps the mirror key set equal
  passes 21/21. The pattern is anchored on or; ⛔ no carrier says so.
• ③C — ⛔ new. A fully amended statement of the rule prepended to data-display.ts passes
  21/21, while the header says a later site 「is covered」. The roster and its exact count are
  complex.ts-scoped by their own text. ⇒ a fifth statement can regrow off-roster and drift at the
  next amendment with nothing red — ⭐ this card's own mechanism, reopened one file over.

Both are prose-and-pin findings, which is now eight of eight nine of nine FAILs this
shift resting there and ⛔ not one on the change's direction.

⚠️ Correction, 2026-09-18T21:45Z, by the seat that wrote the sentence. 「eight」 was wrong when
posted. Measured after the fact by fetching every comment on objectui#9895, objectui#9926 and
objectui#9935 and counting bodies that open ## Contract review and carry a **VERDICT:**
line: 10 records — 9 FAIL, 1 PASS. objectui#9895 carries four FAILs
(5732059417, 5733750066, 5734518119, 5735358922), not the three this seat's own seat post
still recorded; the round-five record was never added to the tally. objectui#9926 carries four
(5734125317, 5734661857, 5735211163, 5736109788); objectui#9935 one FAIL
(5734967234) and the shift's one PASS (5735439084).
⛔ The conclusion is unchanged and the error is the same class this shift has been cataloguing all
day — a closed list published without re-deriving it — this time applied to the seat's own
tally, in a sentence whose whole job was to count. Left visible rather than silently overwritten. ③B is CLOSED — the round did what it reported.

Graded below FAIL, and named so they are not lost

③D: a trailing // on a code line is outside every population the pin enumerates — header claim 2
says 「nothing that matches」 where its assertion says 「no block」, one word wider. ②: the PR body's
banner is bound to 6230bc840 and three of its corrected figures are superseded here (21/21 not 20/20,
six describe blocks not five, 4837 tests not 4836) — vintage-stamped, so ⛔ not misleading, and 「exhaustive」
holds for the head it names.

⭐ ①.4 overturns the body's Clause-②: yes downward: no new exported symbol and no new key on a
published payload, proven by the 65-file .d.ts equality rather than asserted. A yes the review
overturns down ⛔ costs the declarant nothing; a wrong no would have been an auditable false
declaration. The declaration stays right for having been conservative.

⚠️ One reading in ③I this seat owes an answer to

The reviewer re-read the label timeline and every needs:contract-review event is os-sales, with
github-actions[bot] firing as actor on the sibling package: types / tests events — the control
this seat's own retracted objectui#9934 lacked. It also found one event the retraction never
accounted for: labeled 19:44:43Z os-sales, between the implementer's stated 19:44:29Z reading and
its 19:45:14Z comment. The shared identity ⛔ cannot say which seat wrote it. Recorded here rather
than guessed at; it changes nothing about the retraction, which stands.

Carriers

FAIL strips both carriers exactly as PASS does. Card state and assignee ⛔ untouched; the handoff and
the owed changes go on objectui#9684 in the same stroke; re-hung when the head moves.

_{Read at 2026-09-18T21:00Z. Adopting seat: domain:ui#3, seat post objectui#9800.}

---

Generated by Claude Code

[github-actions]

✅ Console Performance Budget

Metric	Value	Budget
Eager closure (gzip, 329 chunks)	3053.3 KB	3104.5 KB
Main entry chunk (gzip)	145.7 KB	350 KB
Entry file	index-DYbGLGEn.js	—
Status	PASS	—

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

Package	Size	Gzipped
app-shell (consoleActionDispatch.js)	0.20KB	0.19KB
app-shell (index.js)	16.69KB	6.21KB
app-shell (runtime-config.js)	20.68KB	7.36KB
app-shell (types.js)	0.01KB	0.04KB
app-shell (urlParams.js)	10.06KB	3.86KB
auth (ActiveOrganizationStorage.js)	25.05KB	9.16KB
auth (AuthContext.js)	0.31KB	0.24KB
auth (AuthGuard.js)	2.07KB	1.00KB
auth (AuthProvider.js)	40.18KB	10.59KB
auth (AuthShell.js)	3.49KB	1.40KB
auth (ForgotPasswordForm.js)	12.21KB	3.45KB
auth (LoginForm.js)	18.15KB	5.39KB
auth (PreviewBanner.js)	0.90KB	0.50KB
auth (RegisterForm.js)	6.65KB	2.22KB
auth (SocialSignInButtons.js)	9.61KB	3.89KB
auth (UserMenu.js)	3.41KB	1.23KB
auth (auth-gate-events.js)	1.29KB	0.66KB
auth (authStyles.js)	5.04KB	1.72KB
auth (createAuthClient.js)	40.21KB	10.80KB
auth (createAuthenticatedFetch.js)	8.46KB	3.43KB
auth (index.js)	3.19KB	1.44KB
auth (invitation-status.js)	1.22KB	0.70KB
auth (org-roles.js)	6.66KB	2.78KB
auth (phone-identifier.js)	1.11KB	0.66KB
auth (types.js)	0.59KB	0.35KB
auth (useAuth.js)	5.30KB	1.02KB
auth (useWorkspaceAdminStatus.js)	11.08KB	4.58KB
collaboration (CommentThread.js)	26.08KB	7.56KB
collaboration (LiveCursors.js)	3.17KB	1.27KB
collaboration (PresenceAvatars.js)	6.49KB	2.64KB
collaboration (PresenceProvider.js)	2.79KB	1.13KB
collaboration (index.js)	1.68KB	0.73KB
collaboration (useCollaborationTranslation.js)	6.05KB	2.52KB
collaboration (useCommentSearch.js)	1.98KB	0.88KB
collaboration (useConflictResolution.js)	7.75KB	1.86KB
collaboration (useMentionNotifications.js)	1.81KB	0.68KB
collaboration (usePresence.js)	6.33KB	1.84KB
collaboration (useRealtimeSubscription.js)	7.91KB	2.01KB
components (index.js)	545.95KB	130.77KB
core (index.js)	8.94KB	3.59KB
create-plugin (index.js)	27.94KB	9.51KB
data-objectstack (index.js)	216.90KB	60.15KB
fields (index.js)	249.62KB	63.02KB
i18n (LocalizationContext.js)	1.76KB	0.96KB
i18n (builtinAggregateLabels.js)	0.86KB	0.49KB
i18n (currency.js)	1.22KB	0.64KB
i18n (fallbackInterpolation.js)	6.25KB	2.77KB
i18n (i18n.js)	8.87KB	3.64KB
i18n (index.js)	5.22KB	2.26KB
i18n (pickLocalized.js)	9.86KB	3.95KB
i18n (provider.js)	32.15KB	10.49KB
i18n (useDisplayLocale.js)	2.85KB	1.45KB
i18n (useObjectLabel.js)	34.34KB	9.17KB
i18n (useSafeTranslation.js)	5.60KB	2.33KB
layout (index.js)	38.83KB	10.95KB
mobile (MobileProvider.js)	0.92KB	0.49KB
mobile (ResponsiveContainer.js)	0.94KB	0.38KB
mobile (breakpoints.js)	1.51KB	0.70KB
mobile (createOfflineDataSource.js)	5.61KB	1.75KB
mobile (index.js)	1.99KB	0.87KB
mobile (offlineQueue.js)	3.91KB	1.35KB
mobile (pwa.js)	0.97KB	0.49KB
mobile (serviceWorker.js)	1.48KB	0.62KB
mobile (serviceWorkerSource.js)	3.41KB	1.48KB
mobile (useBreakpoint.js)	1.54KB	0.65KB
mobile (useGesture.js)	6.96KB	1.98KB
mobile (useOfflineSync.js)	1.99KB	0.72KB
mobile (usePullToRefresh.js)	2.53KB	0.85KB
mobile (useResponsive.js)	0.72KB	0.42KB
mobile (useSpecGesture.js)	5.52KB	2.10KB
mobile (useTouchTarget.js)	1.01KB	0.54KB
permissions (MePermissionsProvider.js)	13.52KB	4.88KB
permissions (PermissionContext.js)	0.31KB	0.25KB
permissions (PermissionGuard.js)	0.89KB	0.45KB
permissions (PermissionProvider.js)	6.24KB	2.16KB
permissions (discardProofCache.js)	1.04KB	0.55KB
permissions (evaluator.js)	8.39KB	3.10KB
permissions (index.js)	0.93KB	0.41KB
permissions (store.js)	0.91KB	0.42KB
permissions (useFieldPermissions.js)	1.28KB	0.53KB
permissions (usePermissions.js)	4.83KB	2.27KB
plugin-ai (index.js)	14.81KB	3.63KB
plugin-calendar (index.js)	49.92KB	14.22KB
plugin-charts (index.js)	71.73KB	20.08KB
plugin-chatbot (index.js)	195.35KB	46.52KB
plugin-dashboard (index.js)	132.96KB	35.17KB
plugin-designer (index.js)	215.94KB	44.33KB
plugin-detail (index.js)	254.30KB	66.21KB
plugin-editor (index.js)	2.23KB	1.05KB
plugin-form (index.js)	137.11KB	34.36KB
plugin-gantt (index.js)	167.62KB	41.26KB
plugin-grid (index.js)	213.44KB	58.21KB
plugin-kanban (index.js)	48.10KB	14.94KB
plugin-list (index.js)	113.42KB	27.95KB
plugin-map (index.js)	21.48KB	6.99KB
plugin-markdown (index.js)	13.88KB	4.80KB
plugin-report (index.js)	43.41KB	11.93KB
plugin-timeline (index.js)	30.07KB	8.74KB
plugin-tree (index.js)	10.58KB	3.72KB
plugin-view (index.js)	85.05KB	21.01KB
providers (DataSourceProvider.js)	0.75KB	0.39KB
providers (MetadataProvider.js)	1.37KB	0.59KB
providers (ThemeProvider.js)	1.90KB	0.85KB
providers (UploadProvider.js)	11.66KB	3.50KB
providers (index.js)	0.45KB	0.23KB
providers (types.js)	0.01KB	0.04KB
react-runtime (index.js)	5.62KB	2.34KB
react (LazyPluginLoader.js)	4.47KB	1.63KB
react (SchemaRenderer.js)	108.85KB	36.01KB
react (data-invalidation.js)	5.05KB	2.08KB
react (index.js)	4.63KB	2.18KB
react (schema-input.js)	4.25KB	2.04KB
react (spec-input.js)	0.20KB	0.18KB
sdui-parser (codegen.js)	6.58KB	2.74KB
sdui-parser (dashboard-widget-options.js)	3.08KB	1.30KB
sdui-parser (index.js)	5.66KB	2.50KB
sdui-parser (input-type.js)	2.84KB	1.40KB
sdui-parser (kanban-quick-add.js)	3.89KB	1.87KB
sdui-parser (parse.js)	25.28KB	7.80KB
sdui-parser (provenance.js)	3.66KB	1.82KB
sdui-parser (types.js)	0.28KB	0.23KB
sdui-parser (validate.js)	14.82KB	4.99KB
types (ai.js)	4.11KB	2.06KB
types (api-types.js)	0.20KB	0.18KB
types (app.js)	2.87KB	1.00KB
types (base.js)	0.20KB	0.18KB
types (blocks.js)	0.20KB	0.18KB
types (complex.js)	2.93KB	1.49KB
types (crud.js)	0.20KB	0.18KB
types (dashboard-filter-alias.js)	6.23KB	2.74KB
types (data-display.js)	3.75KB	1.85KB
types (data-protocol.js)	0.20KB	0.19KB
types (data.js)	0.20KB	0.18KB
types (designer.js)	1.85KB	0.85KB
types (disclosure.js)	0.20KB	0.18KB
types (error-code.js)	1.54KB	0.88KB
types (expression.js)	0.20KB	0.18KB
types (feedback.js)	0.20KB	0.18KB
types (field-types.js)	0.20KB	0.18KB
types (form.js)	0.20KB	0.18KB
types (http-inflight.js)	8.87KB	3.73KB
types (http-retry.js)	4.32KB	2.02KB
types (icon-key-migration.js)	4.26KB	1.63KB
types (index.js)	4.74KB	2.25KB
types (layout.js)	0.20KB	0.18KB
types (managed-by.js)	0.19KB	0.18KB
types (mobile.js)	5.00KB	2.39KB
types (navigation.js)	0.20KB	0.18KB
types (objectql.js)	0.20KB	0.18KB
types (overlay.js)	0.20KB	0.18KB
types (permissions.js)	0.20KB	0.18KB
types (plugin-scope.js)	0.20KB	0.18KB
types (record-components.js)	0.20KB	0.19KB
types (record-semantics.js)	1.28KB	0.67KB
types (registry.js)	0.20KB	0.18KB
types (reports.js)	0.20KB	0.18KB
types (select-option.js)	0.20KB	0.19KB
types (spec-report.js)	5.05KB	1.93KB
types (spec-ui-namespace.js)	0.20KB	0.19KB
types (strict-authoring-face.js)	14.04KB	5.36KB
types (system-fields.js)	3.33KB	1.54KB
types (theme.js)	6.28KB	2.87KB
types (ui-action.js)	8.11KB	3.32KB
types (views.js)	0.20KB	0.18KB
types (widget.js)	0.20KB	0.18KB

Size Limits

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

[os-sales]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: e39b1a8dae42979d297fd9fcaab91338691fc2c9

Read from api.github.com at 2026-09-19T00:10Z: PR #9926 head e39b1a8da…, draft, 7 commits, 3 files (+808 −31), 0 reviews, mergeable: true / behind, auto_merge: null, labels package: types / tests / needs:contract-review, body updated_at 00:06:50Z (the implementer's round-5 re-stamp); card #9684 body and all 14 comments; all 17 PR comments including the four prior records (5734125317, 5734661857, 5735211163, 5736109788) and the round-5 os-dev-report 5737682057. The branch fetched into my own ref equals the API head; main fetched into my own ref is feac43909; merge-base with it and with the API base 0fe29cc23 is fc12bc8c6. main carries two commits under packages/types/src since that base (272a53066 record-components.ts, feac43909 ui-action.ts), none on the three PR files or on scripts/js-comment-mask.mjs. Every measurement below was taken in detached worktrees under my scratchpad: wt at the head (created 00:11:05Z, pnpm install --offline --frozen-lockfile --ignore-scripts exit 0), wt-merge at main with the three PR files restored from the head, wt-gates at the head for the light gates; all three removed with git worktree remove --force at 00:26:28Z and my two refs deleted; the shared checkout /home/user/objectui had 0 dirty lines before (00:10Z) and after (00:26:39Z) and was never edited (HEAD 3e4f6324f, claude/vibrant-galileo-fdqri4). Blobs at head: mobile.ts d75c5adb7, complex.ts e3787c6aa (identical at base), pin 88d332603, changeset f2e2cbdec; base mobile.ts 5e3ad63df. Nothing was posted to GitHub.

① Derived judgments

Enumerated from git diff --stat fc12bc8c6 HEAD: .changeset/9684-discriminator-cited-once.md +93, packages/types/src/__tests__/tombstone-discriminator-agreement-9684.test.ts +602, packages/types/src/mobile.ts +144 −31. Nothing else.

1. mobile.ts: comment-only, no accepted set moved — RIGHT. Of 144 changed lines base→head, 143 begin with // and the one remaining is a blank line (+$ under cat -A); the round-4→round-5 delta has 0 non-comment lines. Item 2 re-derives it.
2. Published .d.ts: no delta — RIGHT, on my own fresh tsc -p packages/types/tsconfig.json builds (dist and tsbuildinfo removed before each) under os-verify-lock (acquired after 0s, held 39s, 00:23:43Z–00:24:22Z). Head: 65 .d.ts, set-hash b578c03369acc604. Base mobile.ts 5e3ad63df swapped in: 65 files, same set-hash, per-file sha256 diff 0 lines. Controls in the same pass: a JSDoc marker in the ChatbotSchema block reaches dist/complex.d.ts 1, dist/mobile.d.ts 0, per-file diff 4 lines against head; a // marker in the mobile.ts preamble is present in src once and reaches dist/mobile.d.ts 0. Restores verified by blob (e3787c6aa, d75c5adb7), empty git diff HEAD, status 0; dist removed afterwards.
3. New test file: not a published surface — RIGHT. packages/types/tsconfig.json excludes **/__tests__/** and **/*.test.ts; tsc -p tsconfig.test.json --noEmit exit 0 (00:23:37Z); check-type-check-coverage exit 0.
4. No new exported symbol, no new key on a published payload — RIGHT. Item 2 is the proof. Clause-② re-derived: no; the declared yes is overturned downward (see ②).
5. Population, my own instrument (pop.mjs: comment spans from scripts/js-comment-mask.mjs, comment-only runs plus trailing one-line blocks, the pin's own markers). Head: 65 files, 4696 blocks, 9 trailing, 4 statements, all complex.ts JSDoc at 165–194, 1026–1097, 1381–1454, 1830–1910, all carrying the precondition, clause sets equal to the roster. Base fc12bc8c6: 4695 blocks, 9 trailing, 8 statements — those four plus mobile.ts 74–120 (no precondition), 122–153, 293–347, 371–400 (no precondition). Merged tree (main feac43909 + PR files): 4701 blocks, 9 trailing, 4 statements. The 9 trailing comments (6 data-protocol.ts, spec-report.ts, zod/base.zod.ts, zod/objectql.zod.ts) state nothing. The body's eight fc12bc8c6 addresses and "9 such comments, none states the rule" — RIGHT.
6. Pin at head 22/22 (00:15:33Z); on the merged tree 22/22 (00:15:36Z). Reverse: mobile.ts at base blob: 8 failed / 14 passed, the precondition and outside-complex.ts assertions naming exactly blocks 74, 122, 293, 371, the four citation assertions and both mobile.ts statement assertions alongside (00:16:14Z); restored d75c5adb7, diff 0, status 0. RIGHT.
7. Ledger cross-checks, each mutation landing proven by a changed blob and restored by git checkout HEAD -- <abs> with blob equality, empty diff and status 0: A1 (, and (iii) …) blob 2a16978c2c0a 22/22 — the blob round 4 and round 5 both printed; A5 (, and (3) …) blob 7ac7178b7d6d 1 failed — the body's blob; B1 amended statement in data-display.ts 1 failed, exactly the outside-complex.ts assertion; my Bz (amended, worded "both prongs … only where a carrier survives", in zod/base.zod.ts) 1 failed, the same; my Bt (un-amended trailing comment on the export interface ActionSchema line in crud.ts) 2 failed (outside + precondition); C1 4 failed; C2 2 failed (both roster assertions); D1 (every statement marker dead) 9 failed / 13 passed including both vacuity controls; S1 (a quotation put back in the header) 1 failed, the self-check; T1 (blank line between the ChatbotSchema JSDoc and its declaration) 1 failed, the target assertion. All RIGHT.
8. Clause-boundary legs at ChatbotSchema, my own wordings — see ③A: one wide-direction falsification.
9. Adjacency legs in complex.ts, nobody's before — see ③B: one population escape.

② Semver level

• Level: empty frontmatter (releases nothing) — matches the diff. check-changeset-presence, -no-major, -fixed, -claims all exit 0 (00:23:11Z–00:23:13Z). ①.2 is what the empty frontmatter asserts. Clause-②: yes on the card → measured no (no accept/reject behaviour moves, no public surface widens; ①.2, ①.3); a conservative yes overturned downward is not a fault.
• Changeset prose vs diff — holds: "eight sites"; "Two sites … MobileResponsiveConfig … MobileComponentConfig"; the Prong 1:/Prong 2: lines verbatim (4 at base, 4 at head, identical multiset after flattening); "the first of those also gains a sentence"; "All 65 … byte-for-byte identical"; the two marker controls; "fails when" × 5 (①.6–①.7); ", and (3) … included" (A5); the four named silent wordings (A1; W-legs). ⛔ Falsified, wide direction: "caught only directly behind or, with nothing but a comma, whitespace and an opening parenthesis allowed in between" — ③A. ⛔ Falsified for one shape: "Its population is every comment in the package" and "fails when … complex.ts states the rule at any number of sites other than the recorded four" — ③B.
• PR body vs diff — holds: 22/22; 8 failed / 14 passed naming 74/122/293/371; 208 files / 4838 tests (under the lock, exit 0, 00:24:01Z); six describe blocks / 22 assertions (3+4+7+1+1+6, read off the file); 9 trailing comments; four complex.ts sites; the position table; A1's and A5's blobs; B1 "exactly one"; C1 4 / C2 2 / D1 9-13; grep -c 1 vs flattened 2 for the finding(types/plugin-chatbot): six ChatbotSchema keys are declared and read by no registration — loading, showAvatars, userAvatar, assistantAvatar, markdown, height #7703 phrase (base and head); lint 0 errors / 290 warnings; NOT GOVERNED — 3 path(s) … 5 governed surface(s). Loose, not load-bearing: "Ablation ledger — 17 legs" while the same paragraph enumerates 10 + 2 + 2 + 1 + the reverse leg + two .d.ts legs = 18; and the pin header's "line end falling inside the two-word name of the carrier condition" holds at one of the two wrapping sites (ChatbotSchema, 1073/1074 SURVIVING | CARRIER) while the other (KanbanColumn.color, 173/174) breaks between "available" and "only" — "reports both as misses" is true for the full-clause probe (raw-line hits at 1407 and 1891 only; flattened 4), safe direction.
• Pending changesets naming mobile.ts: check-changeset-claims exit 0; finding(types/plugin-chatbot): six ChatbotSchema keys are declared and read by no registration — loading, showAvatars, userAvatar, assistantAvatar, markdown, height #7703's phrase 2 at base and 2 at head once flattened.
• Gates in wt-gates (00:23:11Z–00:23:43Z), all exit 0: changeset-presence, no-major, fixed, claims, control-bytes, new-cross-file-line-citations (0 new, report-only), test-path-roots, hand-rolled-comment-mask, type-check-coverage, governed-queue-guard --test on the three paths, eslint on both changed sources, tsc -p tsconfig.test.json --noEmit, pnpm --filter @object-ui/types lint. CI on the head at 00:14:36Z: 36 of 36 check-runs returned (count equals array), 29 success, 3 skipped, 4 test shards in progress — NOT MEASURED as a final state.

③ Boundary flags

A. ⛔ new — the separator sentence names a SET; the pattern enforces a SEQUENCE (STANDS, all four carriers). Appended to the ChatbotSchema statement only: , or ( 3 ) it keeps the mirror key set equal (blob 3dd7dc9432a9) passes 22/22; , or ,3 it keeps … (blob 7858d7d5f1b5) passes 22/22. Both wordings contain nothing but whitespace, an opening parenthesis and a comma between or and the numeral, which the published sentence licenses; the regex \bor,?\s*\(? admits comma, then whitespace, then paren, in that order only — whitespace after the paren, or a comma after whitespace, falls outside it. Failure direction: green. Controls in the same pass, each restored and verified: wrapped , or / (iii) across a line break 1 failed (04b3bc24932c), ; or 3) 1 failed (607141edf530), , OR (III) 1 failed (ad27035ae7dc); , or—(iii), , and third,, , or it keeps (numbers nothing) silent, as declared. Fix: one ordering clause in the four carriers, or widen the separator to a set and re-derive the caught-list. Two observations beside it: , and prong-3 keeps … is silent (ecfac9a227af) — the hyphenated spelling is not "prong 3" and no carrier says so; , or third-party tooling keeps … reds (79a089e72754) — the anchor catches a non-numbering third behind or, the declared spurious-red direction.

B. ⛔ new — two abutting comments are ONE block, so a drifted copy glued to a roster block inherits its precondition and its roster slot (STANDS). In complex.ts: an un-amended // statement of the rule inserted between the ChatbotSchema JSDoc's */ and export interface ChatbotSchema, no blank line (blob 3509c052f2c9): 22/22; an un-amended JSDoc stacked directly above the ChatbotSchema JSDoc, no blank line (8e6ee4ef74d2): 22/22. Control: the same stacked JSDoc with a blank line between (5d283ddf2b01): 4 failed (exact count, roster-names, precondition, shared clauses) — the instrument fires when the blocks are separate. The //-above-JSDoc variant (1dac9b7661d0) reds 2, but only through the two kind === 'jsdoc' assertions because the merged block then starts with //; count and precondition still pass. The pin's own text says "one JSDoc, or one paragraph of // lines" and "A blank line ends a run on purpose — joining across one would let a drifted note inherit a compliant neighbour's precondition and pass" — the code joins across */→// and */→/** with no blank line, which is exactly that inheritance; a human reads five statements and the pin counts four. Measured in the package today: */→// with no blank 0, */→/** with no blank 0, //→/** with no blank 5 (index.ts 1085, 1129, 1170, 1224, 1428) — the two silent shapes are not in the tree, the present one is caught. Fix: end a run at a */ line and start one at a kind change — the code moved to the prose, the same move ③D took.

C. Standing A (silent-set summary) — CLOSED for every wording it was raised on; the replacement sentence carries finding A above.

D. Standing C / B1 (off-roster statement) — CLOSED. B1 reds on exactly the new assertion; Bz in a subdirectory the same; D1 shows the zero is not a dead-marker zero.

E. Standing ③D (trailing comments) — CLOSED. C1 4, C2 2, Bt 2; 9 trailing at head, none states the rule.

F. ③E (whole-type-name clause roster-only) — observation, left alone, NOT MEASURED this round.

G. Merged-tree reading, not a finding: the PR is behind; on main + the three PR files the pin is 22/22 and the population is 4 — recorded because the queue rebuilds on main.

H. Label event, recorded, not concluded: needs:contract-review unlabeled 21:00:32Z and labeled 00:06:50Z, both os-sales; the labeled event is the same second as the body's updated_at. The body says this round wrote none; the shared identity cannot say which seat did.

I. NOT MEASURED: CI final state (four shards running at 00:14:36Z); vite build + dist completeness (the body's 130 files — I ran tsc only); a third head re-build (my third build was the marker control); legs A2–A4, A6–A10 and B2 individually (covered by A1, A5, the W-legs and Bt in their classes); the card's reading on dce4a74a1; a machine scan of the body for closing keywords (read by eye: "Part of #9684" only).

J. Served tier. My transcript is the one file in subagents/ carrying r5-marker-1789776608-23093: agent-a97ff63c6ca61fd6d.jsonl — 86 assistant requests carry a harness model stamp at 00:26:16Z, all claude-fable-5-1, zero other values. Discriminating controls on the same probe: agent-ad385c5d27f779b84.jsonl 375/375 claude-opus-5, agent-a44043eb5c3272baf.jsonl 832/832 claude-opus-5. CONTRACT_REVIEW_TIER is declared 'claude-fable-5-1' at its export const in /home/user/objectstack/scripts/pm/dispatch-gates.mjs; the measured value matches it.

Implemented-by: claude/issue-9684-retire-remove-discriminator-single-source
Reviewed-by: session_01Xm4WFhEe5mwcgyqHjxR2hn

VERDICT: FAIL

In plain prose: I reviewed head e39b1a8dae42979d297fd9fcaab91338691fc2c9, unchanged through my last read, from my own detached worktrees, and removed them clean. The round did everything it reported, and I could reproduce every figure it printed — the pin, the reverse leg, the full suite, the 65-file .d.ts equality with both controls, the population at base and head, the ledger's blobs for A1 and A5, and the exact failure sets for B1, C1, C2 and D1 — so both standing findings and ③D are closed, and the direction is right as it has been every round. The FAIL rests on two things nobody had tried, both in the silent direction: the new separator sentence licenses or ( 3 ) and or ,3, which the pattern does not catch, because it names a set of characters where the regex requires an order; and the block definition merges abutting comments, so an un-amended copy glued directly under (or as a JSDoc directly above) a roster block passes at 22/22 while the same copy with a blank line reds 4. Neither escaping shape exists in the tree today, and both fixes are small — an ordering clause in four places, and ending a comment run at a kind boundary — but each is a stated reach the pin does not have, which is the standard this review is held to.

---

Generated by Claude Code

[github-actions]

✅ Console Performance Budget

Metric	Value	Budget
Eager closure (gzip, 329 chunks)	3053.5 KB	3104.5 KB
Main entry chunk (gzip)	145.7 KB	350 KB
Entry file	index-C3YkV3C4.js	—
Status	PASS	—

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

Package	Size	Gzipped
app-shell (consoleActionDispatch.js)	0.20KB	0.19KB
app-shell (index.js)	16.69KB	6.21KB
app-shell (runtime-config.js)	20.68KB	7.36KB
app-shell (types.js)	0.01KB	0.04KB
app-shell (urlParams.js)	10.06KB	3.86KB
auth (ActiveOrganizationStorage.js)	25.05KB	9.16KB
auth (AuthContext.js)	0.31KB	0.24KB
auth (AuthGuard.js)	2.07KB	1.00KB
auth (AuthProvider.js)	40.18KB	10.59KB
auth (AuthShell.js)	3.49KB	1.40KB
auth (ForgotPasswordForm.js)	12.21KB	3.45KB
auth (LoginForm.js)	18.15KB	5.39KB
auth (PreviewBanner.js)	0.90KB	0.50KB
auth (RegisterForm.js)	6.65KB	2.22KB
auth (SocialSignInButtons.js)	9.61KB	3.89KB
auth (UserMenu.js)	3.41KB	1.23KB
auth (auth-gate-events.js)	1.29KB	0.66KB
auth (authStyles.js)	5.04KB	1.72KB
auth (createAuthClient.js)	40.21KB	10.80KB
auth (createAuthenticatedFetch.js)	8.46KB	3.43KB
auth (index.js)	3.19KB	1.44KB
auth (invitation-status.js)	1.22KB	0.70KB
auth (org-roles.js)	6.66KB	2.78KB
auth (phone-identifier.js)	1.11KB	0.66KB
auth (types.js)	0.59KB	0.35KB
auth (useAuth.js)	5.30KB	1.02KB
auth (useWorkspaceAdminStatus.js)	11.08KB	4.58KB
collaboration (CommentThread.js)	26.08KB	7.56KB
collaboration (LiveCursors.js)	3.17KB	1.27KB
collaboration (PresenceAvatars.js)	6.49KB	2.64KB
collaboration (PresenceProvider.js)	2.79KB	1.13KB
collaboration (index.js)	1.68KB	0.73KB
collaboration (useCollaborationTranslation.js)	6.05KB	2.52KB
collaboration (useCommentSearch.js)	1.98KB	0.88KB
collaboration (useConflictResolution.js)	7.75KB	1.86KB
collaboration (useMentionNotifications.js)	1.81KB	0.68KB
collaboration (usePresence.js)	6.33KB	1.84KB
collaboration (useRealtimeSubscription.js)	7.91KB	2.01KB
components (index.js)	545.95KB	130.77KB
core (index.js)	8.94KB	3.59KB
create-plugin (index.js)	27.94KB	9.51KB
data-objectstack (index.js)	216.90KB	60.15KB
fields (index.js)	249.62KB	63.02KB
i18n (LocalizationContext.js)	1.76KB	0.96KB
i18n (builtinAggregateLabels.js)	0.86KB	0.49KB
i18n (currency.js)	1.22KB	0.64KB
i18n (fallbackInterpolation.js)	6.25KB	2.77KB
i18n (i18n.js)	8.87KB	3.64KB
i18n (index.js)	5.22KB	2.26KB
i18n (pickLocalized.js)	9.86KB	3.95KB
i18n (provider.js)	32.15KB	10.49KB
i18n (useDisplayLocale.js)	2.85KB	1.45KB
i18n (useObjectLabel.js)	34.34KB	9.17KB
i18n (useSafeTranslation.js)	5.60KB	2.33KB
layout (index.js)	38.83KB	10.95KB
mobile (MobileProvider.js)	0.92KB	0.49KB
mobile (ResponsiveContainer.js)	0.94KB	0.38KB
mobile (breakpoints.js)	1.51KB	0.70KB
mobile (createOfflineDataSource.js)	5.61KB	1.75KB
mobile (index.js)	1.99KB	0.87KB
mobile (offlineQueue.js)	3.91KB	1.35KB
mobile (pwa.js)	0.97KB	0.49KB
mobile (serviceWorker.js)	1.48KB	0.62KB
mobile (serviceWorkerSource.js)	3.41KB	1.48KB
mobile (useBreakpoint.js)	1.54KB	0.65KB
mobile (useGesture.js)	6.96KB	1.98KB
mobile (useOfflineSync.js)	1.99KB	0.72KB
mobile (usePullToRefresh.js)	2.53KB	0.85KB
mobile (useResponsive.js)	0.72KB	0.42KB
mobile (useSpecGesture.js)	5.52KB	2.10KB
mobile (useTouchTarget.js)	1.01KB	0.54KB
permissions (MePermissionsProvider.js)	13.52KB	4.88KB
permissions (PermissionContext.js)	0.31KB	0.25KB
permissions (PermissionGuard.js)	0.89KB	0.45KB
permissions (PermissionProvider.js)	6.24KB	2.16KB
permissions (discardProofCache.js)	1.04KB	0.55KB
permissions (evaluator.js)	8.39KB	3.10KB
permissions (index.js)	0.93KB	0.41KB
permissions (store.js)	0.91KB	0.42KB
permissions (useFieldPermissions.js)	1.28KB	0.53KB
permissions (usePermissions.js)	4.83KB	2.27KB
plugin-ai (index.js)	14.81KB	3.63KB
plugin-calendar (index.js)	49.92KB	14.22KB
plugin-charts (index.js)	71.73KB	20.08KB
plugin-chatbot (index.js)	195.35KB	46.52KB
plugin-dashboard (index.js)	132.96KB	35.17KB
plugin-designer (index.js)	215.94KB	44.33KB
plugin-detail (index.js)	254.30KB	66.21KB
plugin-editor (index.js)	2.23KB	1.05KB
plugin-form (index.js)	137.49KB	34.64KB
plugin-gantt (index.js)	167.62KB	41.26KB
plugin-grid (index.js)	213.44KB	58.21KB
plugin-kanban (index.js)	48.10KB	14.94KB
plugin-list (index.js)	113.42KB	27.95KB
plugin-map (index.js)	21.48KB	6.99KB
plugin-markdown (index.js)	13.88KB	4.80KB
plugin-report (index.js)	43.41KB	11.93KB
plugin-timeline (index.js)	30.07KB	8.74KB
plugin-tree (index.js)	10.58KB	3.72KB
plugin-view (index.js)	85.05KB	21.01KB
providers (DataSourceProvider.js)	0.75KB	0.39KB
providers (MetadataProvider.js)	1.37KB	0.59KB
providers (ThemeProvider.js)	1.90KB	0.85KB
providers (UploadProvider.js)	11.66KB	3.50KB
providers (index.js)	0.45KB	0.23KB
providers (types.js)	0.01KB	0.04KB
react-runtime (index.js)	5.62KB	2.34KB
react (LazyPluginLoader.js)	4.47KB	1.63KB
react (SchemaRenderer.js)	108.85KB	36.01KB
react (data-invalidation.js)	5.05KB	2.08KB
react (index.js)	4.63KB	2.18KB
react (schema-input.js)	4.25KB	2.04KB
react (spec-input.js)	0.20KB	0.18KB
sdui-parser (codegen.js)	6.58KB	2.74KB
sdui-parser (dashboard-widget-options.js)	3.08KB	1.30KB
sdui-parser (index.js)	5.66KB	2.50KB
sdui-parser (input-type.js)	2.84KB	1.40KB
sdui-parser (kanban-quick-add.js)	3.89KB	1.87KB
sdui-parser (parse.js)	25.28KB	7.80KB
sdui-parser (provenance.js)	3.66KB	1.82KB
sdui-parser (types.js)	0.28KB	0.23KB
sdui-parser (validate.js)	14.82KB	4.99KB
types (ai.js)	4.11KB	2.06KB
types (api-types.js)	0.20KB	0.18KB
types (app.js)	2.87KB	1.00KB
types (base.js)	0.20KB	0.18KB
types (blocks.js)	0.20KB	0.18KB
types (complex.js)	2.93KB	1.49KB
types (crud.js)	0.20KB	0.18KB
types (dashboard-filter-alias.js)	6.23KB	2.74KB
types (data-display.js)	3.75KB	1.85KB
types (data-protocol.js)	0.20KB	0.19KB
types (data.js)	0.20KB	0.18KB
types (designer.js)	1.85KB	0.85KB
types (disclosure.js)	0.20KB	0.18KB
types (error-code.js)	1.54KB	0.88KB
types (expression.js)	0.20KB	0.18KB
types (feedback.js)	0.20KB	0.18KB
types (field-types.js)	0.20KB	0.18KB
types (form.js)	0.20KB	0.18KB
types (http-inflight.js)	8.87KB	3.73KB
types (http-retry.js)	4.32KB	2.02KB
types (icon-key-migration.js)	4.26KB	1.63KB
types (index.js)	4.74KB	2.25KB
types (layout.js)	0.20KB	0.18KB
types (managed-by.js)	0.19KB	0.18KB
types (mobile.js)	5.00KB	2.39KB
types (navigation.js)	0.20KB	0.18KB
types (objectql.js)	0.20KB	0.18KB
types (overlay.js)	0.20KB	0.18KB
types (permissions.js)	0.20KB	0.18KB
types (plugin-scope.js)	0.20KB	0.18KB
types (record-components.js)	0.20KB	0.19KB
types (record-semantics.js)	1.28KB	0.67KB
types (registry.js)	0.20KB	0.18KB
types (reports.js)	0.20KB	0.18KB
types (select-option.js)	0.20KB	0.19KB
types (spec-report.js)	5.05KB	1.93KB
types (spec-ui-namespace.js)	0.20KB	0.19KB
types (strict-authoring-face.js)	14.04KB	5.36KB
types (system-fields.js)	3.33KB	1.54KB
types (theme.js)	6.28KB	2.87KB
types (ui-action.js)	8.11KB	3.32KB
types (views.js)	0.20KB	0.18KB
types (widget.js)	0.20KB	0.18KB

Size Limits

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

[os-tesla]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: baa386c4bf900c03dc1879531f145ecf1b97ceb9

① Derived judgments

Every reading below is my own, taken on the head blob set against its merge-base with origin/main, fc12bc8c67494a0e3052d91a12dcede8a2fcc60b (the diffstat between them is exactly the three files named in the brief: 914 added, 31 deleted). This container had no node_modules in either checkout, so the instruments are named with their provenance: the repo's own scripts/js-comment-mask.mjs scanner (dependency-free, run in place), the globally installed tsc 6.0.2 (the repo pins ^6.0.3), vitest 5.0.1 fetched by npx into a scratch tree (the repo pins ^4.1.10), and the repo's own gate scripts run inside a --shared clone of the checkout at the head sha. Nothing under /home/user/objectui was written.

• mobile.ts code — no accepted set moves; judged right. Masking every comment span with the shared scanner and diffing the residue: base and head are identical (101 code lines each, DIFF EMPTY). Same-command controls: one code token changed in base (GestureType renamed) makes the masked residue differ; one comment token changed leaves it identical. Every one of the 159 changed diff lines in mobile.ts is a // line except one added blank line (control: the changeset diff's 113 changed lines are all non-//). The 18 export lines are identical at base and head, the single runtime export SPEC_GESTURE_TYPES included. No type, member, union, key or export is added, removed, renamed or retyped.
• Published declarations — no published surface moves; judged right. Running tsc -p packages/types/tsconfig.json --emitDeclarationOnly over the package source twice, once with the head mobile.ts and once with the base blob 5e3ad63df swapped in: 65 .d.ts files each, per-file sha256 tables identical (DIFF EMPTY: 65 files identical). Same-command controls that fire: a JSDoc marker placed directly above export interface GestureContext reaches mobile.d.ts once and moves the table by 2 lines; a // marker in the same place reaches it 0 times and moves the table by 0 lines. A six-form micro-probe agrees: only /** */ comments (attached or on a member) survive declaration emit, while // and plain /* */ comments at module scope, attached to a declaration, or inside an interface reach the .d.ts 0 times (each marker present once in source as the planting control). The emit ran without node_modules, so every leg carried the same 124 unresolved-module errors; emit is unaffected and the controls show the instrument fires. The new preamble and rewritten notes are all // comments, and mobile.ts carries no legal-comment marker (@license, @preserve, //!, /*!: 0, against 2 in the pin's license header), so the vite build JS output has nothing to keep either; that JS emit was not rebuilt here and is inferred from those two readings plus the identical code residue.
• complex.ts — untouched; judged right to leave. Not in the diffstat. Its four JSDoc statements are the published half the card leaves open, and the head does not move a byte of it. The four roster anchors (RETIRED with the declarative face (objectui#7664, Chatbot component — the authoring face, ADR-0049 RETIREMENT TOMBSTONE — displayMode, ADR-0049 RETIREMENT TOMBSTONE — triggerIcon) each occur exactly once in the head complex.ts; the ChatbotSchema JSDoc directly above export interface ChatbotSchema states the rule with the surviving-carrier precondition.
• The new pin tombstone-discriminator-agreement-9684.test.ts — not a published surface; judged right. It sits under __tests__, which packages/types/tsconfig.json excludes by directory, and package.json files publishes only dist, README.md, CHANGELOG.md, LICENSE. Run in a scratch tree of the head's packages/types/src plus scripts/js-comment-mask.mjs: 22 passed of 22. Reverse leg with the base mobile.ts blob swapped in (on-disk blob 5e3ad63df confirmed): 8 failed, 14 passed; the outside-complex.ts assertion names four mobile.ts blocks, and the precondition assertion names exactly two, the ones opening the MobileResponsiveConfig and MobileComponentConfig retirement notes. My direct read of those two base notes agrees: 0 occurrences of "surviving carrier" in each, with the MobileOverrides and GestureConfig notes as lit controls at 1 each, and the MobileComponentConfig note carrying the paraphrase "no surviving object" once. So the base population was eight statements with two un-amended, both in mobile.ts, as the PR says and the card did not. The widened a-prong-beyond-the-two separator ([\s,(]* replacing ,?\s*\(?) is a tightening of an internal guard that publishes nothing, not a change to any accepted set.
• Queue-rebuild preview — judged clean. Seven commits on origin/main since the base touch packages/types/src; 0 touch mobile.ts or complex.ts (the seven are listed by the same git log). The pin run over origin/main 078f2e4f7's packages/types/src with this branch's mobile.ts and pin overlaid: 22 passed of 22.
• Prose semantics inside // comments — judged right, unpublished. The MobileResponsiveConfig note gains the precondition it lacked; the MobileOverrides, GestureConfig and MobileComponentConfig notes replace their restatement with a citation of the ChatbotSchema statement while keeping every Prong 1: / Prong 2: measurement and reopen condition; "no surviving object" becomes the rule's own "no surviving carrier". A pure re-citation with no semantic move, proven from the diff and the emit rather than from the title. One observation, not a finding: the new ~90-line preamble narrates review-round history ("the fourth in-seat review's", "the fifth review's") in shipped source comments; it is //, reaches no consumer, and states per AGENTS.md 完善设计器的每一个细节 #9 which of its readings nothing re-derives.

② Semver level

The changeset .changeset/9684-discriminator-cited-once.md has an empty frontmatter (two --- lines, verified with cat -A) and declares, flattened across its line wrap, "Nothing publishes" (count 1; control: the precedent 7678 changeset says "releases nothing" once). No level is declared, which is the repo's first-class "releases nothing" form; scripts/check-changeset-presence.mjs run in the clone at the head accepts it verbatim ("declared as releasing nothing, which is the explicit exemption"), and check-changeset-no-major passes. That matches ①: zero accepted-set moves and zero published bytes, so no release is the correct declaration; patch would over-declare and minor/major would be wrong.

The Clause-② declaration is yes on the claim comment and in the PR body (API read of the body: one line, Clause-②: yes). ① finds no widening, no narrowing and no new exported symbol, so this review overturns it downward: the change is Clause-②: no in substance. Per references/contract-review.md the claim-time yes is the conservative gate direction and being overturned by the in-seat review is by design and not a seat fault; the claim itself gave the reason (an implementation hanging the shared block on a new exported symbol could not be ruled out at claim time, and was not the implementation taken). One cosmetic inconsistency, not a gate input: the changeset body carries no Clause-②: line (0; the same-class precedent 7678 carries 1), while the gate reads the claim comment's line-start key; only 8 of the 1729 pending changesets carry such a line, so its absence breaks no convention.

③ Boundary flags

• Decision box: convert the four complex.ts statements to citations, or leave them. Escalated to the maintainer as objectui#9930 (open, needs-user-decision), which also carries the roster asymmetries (displayMode without the whole-type-name clause, triggerIcon without either prong clause) and the question of un-skipping __tests__. Answer: correctly escalated. This PR takes only the half that moves zero published bytes, nothing in ① depends on the outcome, and every option in [decision] the retire-vs-remove discriminator is stated at four sites in published JSDoc — convert them to cite one block, leave them, or make the pin enforce their agreement #9930 that edits published JSDoc would need its own real changeset. Not a blocker.
• The canonical block does not say it is canonical. Deferred to [decision] the retire-vs-remove discriminator is stated at four sites in published JSDoc — convert them to cite one block, leave them, or make the pin enforce their agreement #9930 because naming it edits ChatbotSchema's JSDoc and would move .d.ts bytes. Answer: right to defer; doing it here would have made ② false.
• The pin's declared silent set (a clause worded past every marker; , and (iii) …, , or else (iii) …, ; and, third, …, , or: (iii) …, an em-dash join, prong-3, and any wording past STATES_THE_RULE), and the declared spurious-red direction (, or third-party tooling reds). Answer: accepted as the stated reach of an internal guard that is not a published contract; a spurious red sends a reader to four blocks, a spurious green is the defect the card was filed about, so the direction is the right one. I did not re-run the 21 separator legs; the pin's regex literals were read and the head and reverse legs reproduce. Observation carried to the next editor: the caught-list is restated in the pin header, the CLAUSES comment, the mobile.ts preamble and the changeset, and each says to re-derive it from the pattern when the pattern moves.
• Round 5's "pattern unchanged" reasoning reversed in round 6. Answer: the reversal is right. Widening a test regex creates no authorable key, exported symbol or published spelling; it is a tightening and belongs under the "loud refusal beats silent tolerance" axis, not the "no new capability surface" axis.
• Standing-rules conflict: the dispatch names the PR-body re-stamp as owed, the os-dev clause forbids a body PATCH. Escalated as objectui#9976 (open, needs-user-decision). Answer: a process question, not a contract one; correctly escalated, and the body PATCH is rightly not rolled back since the body at this head is one vintage bound to baa386c4b.
• "NOT MEASURED: CI's final state on this head." Measured here from the head's check runs: 36 returned, 33 success, 3 skipped (dependabot, Test (coverage), Test (coverage shard)), 0 failure, 0 pending; Governed Surface Queue Guard, Line Citation Gate, Changeset Declaration, Type Check and Test (shard 1/4) through (shard 4/4) all success. The full packages/types suite was not re-run in this container; CI's shards are the reading for it.
• check:changeset-claims: two pending changesets name mobile.ts. 7580-responsive-vocabulary-rehome.md claims BreakpointName is declared in mobile.ts: true at head (export type BreakpointName = … present). 7703-chatbot-dark-keys-retired.md quotes "no silent-strip hazard for prong 2 to guard" from mobile.ts: present twice at head when line wrapping is flattened (1 by line-based grep -c, one occurrence wraps), the same two as at base; control: an unwrapped once-only phrase reads 1 by both methods. On current origin/main 7703 still names mobile.ts and still quotes the phrase after PR docs(changeset): four pending bodies cite the retire-vs-remove discriminator instead of restating it #9970 merged (which closed objectui#9932), so neither claim is born false by this diff.
• Governed surface. node scripts/check-governed-queue-guard.mjs --test on the three changed paths: NOT GOVERNED, exit 0; control: the same command on AGENTS.md reports GOVERNED, exit 3. The ordinary review-and-queue route applies once the seat clears the carriers.
• Cross-file line citations (AGENTS.md [WIP] Update documentation for project #11). The repo's own gate check-new-cross-file-line-citations.mjs --base fc12bc8c6 run in the clone at the head: 0 new citations, exit 0; control: the same gate against origin/main~300 reports 28 new citations in that span, so the instrument is live. A crude file.ext:NN scan of the three files: 0, 0, 0, with scripts/body-dialect-census.mjs as the lit control at 2. The pin keys its roster by anchor text and prints block lines only at run time, which is the permitted computed form.
• Closing keywords. The PR body has 0 closing keywords immediately before an issue reference (control: fixes #1 and closes objectui#2 scores 2), so Part of #9684 leaves the card open for [decision] the retire-vs-remove discriminator is stated at four sites in published JSDoc — convert them to cite one block, leave them, or make the pin enforce their agreement #9930 as intended; the body carries 0 angle-bracket fragments. needs:contract-review is present on the PR at this head; the body's "this round wrote no label" is consistent with the seat's own record of hanging it.
• "noted, not filed" items in the body. The round-5 handoff's abutting-comment census that did not reproduce (3/13/11 versus 0/0/5) was not independently re-derived here and is not load-bearing for ①②: what matters, the pin's exact-four roster at head and its 8/14 reverse reading, reproduces. The card's stale position table (measured on an open PR's head that merged thirty minutes later) is a card-side lesson with no action on this PR. AIFormAssistSchema.formId in ai.ts citing "PRONG 2 of the discriminator" stays outside STATES_THE_RULE by construction and the pin passes with ai.ts in the population, so the precedent is real and untouched. ObjectKanbanSchema.allowCollapse in objectql.ts argues a different mechanism and is correctly not counted as a ninth site. The round-4 review's suggested wording being one step wide is historical and already corrected by the seat on the card.
• Card body "Not measured" items. Whether the same N-copies shape exists for other rules in packages/types remains unmeasured and stays on the card; it is outside this PR and does not gate it. Whether the un-amended GestureConfig note misled anyone is moot, that note having been amended by 5f8190c8c before this branch's base. What a shared block costs a .d.ts consumer is now measured, by the PR and independently above: JSDoc form reaches the declarations, // form does not, which is exactly why converting mobile.ts cost consumers nothing and converting complex.ts is [decision] the retire-vs-remove discriminator is stated at four sites in published JSDoc — convert them to cite one block, leave them, or make the pin enforce their agreement #9930's trade.
• The withdrawn labeler-wipe claim (objectui#9934, closed not-a-defect). Not a flag on this head; the round-6 body does not carry it.

Implemented-by: claude/issue-9684-retire-remove-discriminator-single-source
Reviewed-by: session_018HrVaotisyhgmot9o2MLRq

VERDICT: PASS

---

Generated by Claude Code

[os-tesla]

Clause-② gate CLEARED on a PASS — provenance. domain:ui#2 execution seat (PM session session_018HrVaotisyhgmot9o2MLRq), 2026-09-20T09:29Z, under the maintainer's direct instruction 「#9926 / #9993 你可以执行契约复审」.

• Record: comment 5748954714 on this PR — a ## Contract review naming head baa386c4bf900c03dc1879531f145ecf1b97ceb9, VERDICT: PASS.
• Both carriers cleared in one stroke: needs:contract-review removed from this PR at 2026-09-20T09:29:08Z and from card objectui#9684 at 2026-09-20T09:29:09Z — four-step writes, each read back.
• Machine check: PM_SWEEP_REPO=objectstack-ai/objectui node scripts/pm/check-clause2-carriers.mjs --pair 9926 → exit 0, both carriers agreeing and the record naming this head.
• Tier: ⛔ not a self-review. CONTRACT_REVIEW_TIER is claude-fable-5-1 and this seat is served Opus, so per the fuse the review was rendered by an isolated at-tier subagent, briefed adversarially and fed only this card, its rulings and the PR — ⛔ never this seat's conclusions. Adopted verbatim.

Two things in that record the next reader should not skim past

① The review overturns the claim's Clause-②: yes downward to no in substance — it finds no widening, no narrowing and no new exported symbol, and proves it two ways: the comment-masked code residue is byte-identical between base and head (101 code lines each), and the emitted .d.ts set is identical file-for-file (65 files, per-file sha256 table DIFF EMPTY), with planted JSDoc and // markers as the controls that fire and do not. Per references/contract-review.md a claim-time yes overturned by the in-seat review is by design and ⛔ not a seat fault.

② The changeset declares 「Nothing publishes」 with empty frontmatter — the repo's first-class releases-nothing form, which check-changeset-presence.mjs accepts verbatim. That matches ① exactly: patch would over-declare.

⇒ Flipping ready and enqueuing. ⛔ No approval review submitted by this seat, ⛔ nothing merged by hand. The two questions the record names as escalated (objectui#9930, objectui#9976) stay open and ⛔ do not gate this PR.

---

Generated by Claude Code