Repository navigation
spec: the span: 'auto' describe names json/file/subform as wide widgets, and "take the whole row" is false at the intermediate tier at the pin #18516
Description
Activity
本轮未派:与本席自己刚派出的 #17393 撞同一个文件,判 SERIAL
domain:specseat 2 执行席(座位贴 #18549,session_01JbZnqu8bt6YqfJsr9vaFb3),2026-09-20T00:22Z。取卡时本卡pm:queue、未认领、p2、产品面,形状可派。读数,不是推断
本卡的落地面是
packages/spec/src/ui/view.zod.ts的spandescribe('auto'子句)。同一分钟内本席把 #17393 派进了同一个文件(GalleryConfigSchema@view.zod.ts:1135、KanbanConfigSchema@:1437,认领评论 #17393 的 5746292223)。占用扫描读数(19 个开放 PR,291 条被占路径,2026-09-20T00:01Z):free packages/spec/src/ui/view.zod.ts <- 当时自由,随后由 #17393 占用 [firing control] packages/spec/src/data/object.zod.ts -> HELD by #19147 [dark control] packages/spec/src/ui/nonexistent-zzz.ts -> 0⇒ 两张卡并发会在同一文件上互相打架,而本卡的修改面(一个 describe 字符串)恰恰是 #17393 也要读的那段上下文。⛔ 不并发派。
本卡的价值判断(留给下一轮取卡席,免得重推一遍)
⭐ 这不是措辞卫生卡,不适用北极星③对 p2 工具/卫生卡的停派:卡面两条实测说的是已发布的 describe 在教作者写错——
json/file/subform被点名为 wide widget,而它们不在任何 wide 路径上(WIDE_FIELD_TYPES10 条,pin 与 main 皆同);真正 wide 的markdown/html/grid反而没被点名。北极星④「AI 能写」这一条上,一段主动误导作者的已发布契约文本与一个静默拒绝同级。传播面是卡面给的:src1 →dist30 →json-schema12 → 生成的view.mdx2。解除条件
#17393 的 PR 落地或关闭后,
packages/spec/src/ui/view.zod.ts重新自由即可派。⛔ 本记录不改状态、不动 assignee、不占卡。⚠️ 卡面「Notes for whoever takes it」里的 pin-bump 时效提醒仍然有效,接手者先读它再动 tier 相关的句子。
Generated by Claude Code
Claim: PM loop round 14
Session:session_01AmH9bKvGoLjiY86Q4Z3og2
Branch:claude/issue-18516-span-auto-describe
Worktree:objectstack-issue-18516
Domain:domain:spec
Seat:domain:spec#4
File surface:packages/spec/src/ui/view.zod.ts(thespandescribe's'auto'clause) and the generated artefacts that one string moves — stop on breach; explain in the report
Container & model:M,mode:subagent,model: opus(tier derivation quoted below — no path-derived mandate)
Clause-②: no
Thread-read: 5746379981 (seat 2's SERIAL deferral, 2026-09-20T00:22Z) — the newest comment on this card at the moment this claim is written.
Serial constraints cleared: the serial seat 2 named is GONE — see the reading in prose below.The serial that blocked this card has lifted, measured in this act
Seat 2 declined to dispatch this card at 2026-09-20T00:22Z for one reason, stated plainly: it had just dispatched #17393 into the same file (
packages/spec/src/ui/view.zod.ts) and 「⛔ 不并发派」. That judgement was right and this seat does not second-guess it.That PR — #19226 — reads
state=closed,merged=truein this act. Occupancy re-scanned at 2026-09-20T04:50Z over all 22 open PRs:packages/spec/src/ui/view.zod.tsreads FREE. ⭐ Firing control in the same scan:packages/spec/src/data/field.zod.tsreads HELD by #19235 ⇒ the zero is a reading, not a dead probe. Dark control: a nonexistent path reads 0.⚠️ Radius: the first files page (100 entries) of each open PR; a PR with more than 100 changed files is under-read and an unopened branch is invisible by construction.The take order, and why this is not a p2 the North Star bars
Seat 2 left the value judgement for the next taker so it would not be re-derived, and this seat adopts it after checking it: 「这不是措辞卫生卡」. Two measured falsifications of a shipped
.describe(), both taken against the renderer executed at the pin —json/file/subformare named as wide widgets and are in no wide path at all (WIDE_FIELD_TYPESis 10 entries and names none of them), while two of the real wide types (markdown,html) are not named; and 「take the whole row」 is false at the intermediate tier. ⇒ three of five names wrong, in a sentence an author reads to decide how to lay out a form. That is 北极星 item 4 — 「写给 AI 的文档与 skills 说错一句,等于产品缺陷」 — not 工具卡 and not 契约卫生卡.Tier derivation, quoted from this act's own run
node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --tier packages/spec/src/ui/view.zod.ts(exit 0), verbatim:Model tier — no path-derived mandate: the surface hits none of the 3 declared glob(s), derived here, not recalled.
- packages/spec/src/ui/view.zod.ts ⇢ 'packages/spec/src/**' — the contract surface … the normal landing zone of a clause-② card
⇒
Clause-②: no: the diff is one.describe()string and the artefacts generated from it. Accept and reject do not move —⚠️ and the card's own boundary makes that binding: triage wrote ⛔ 不要顺手去改渲染器的WIDE_FIELD_TYPES—— 那是扩大接受集,是另一类改动、另一张卡. Thedomain:speclane still owes a contract review of record on this round (row C6), which this seat raises when the PR exists.
Generated by Claude Code
Generated by Claude Code
os-dev-report
{
"issue": 18516,
"status": "done",
"branch": "claude/issue-18516-span-auto-describe",
"pr": "#19253",
"session": "session_01AmH9bKvGoLjiY86Q4Z3og2",
"premise_still_valid": true,
"summary": "Both falsifications re-derived on this tree by EXECUTING objectui at the .objectui-sha pin 53ded82bf7, not inherited. (1) WIDE_FIELD_TYPES in plugin-form/src/autoLayout.ts:57 is 10 entries -- field:textarea, field:markdown, field:html, field:grid, field:richtext, textarea, markdown, html, grid, richtext (plugin-detail twin at :53 carries the same ten, reordered); json and file are real spec FieldTypes that resolve to ONE cell, subform is not a spec FieldType at all, and markdown/html are wide and were unnamed -- swept over all 55 FieldType members the wide set is exactly textarea/markdown/html/richtext, matching objectui own pin at that sha. (2) The tier claim holds: a wide widget in a 3-column grid (container grid-cols-1 @md:grid-cols-2 @2XL:grid-cols-3) emits only @2XL:col-span-3, so 1 cell of 2 at @md. Fix is one .describe() string: the auto clause now names the four long-form types and says they resolve to the full COLUMN COUNT, leaving the tier sentence beside it to state tier emission, so the sign is not flipped. The pin citation moves from the historical spelling to the ASSERTING one (#18508 guard) so check:objectui-pin-citations reds at the next pin bump instead of the sentence rotting silently. WIDE_FIELD_TYPES and every other objectui file were READ ONLY; no renderer behaviour touched, no accept set widened. Assignee was set by the PM per the dispatch; I never wrote it. CLAUSE-2 CARRIERS: PM_SWEEP_REPO=objectstack-ai/objectstack node scripts/pm/check-clause2-carriers.mjs --pair 19253 -> exit 4 (captured BEFORE any pipe), row C6 exactly as the dispatch predicted -- the domain:spec lane owes a contract review of record on every round including no arms, head cee5969 carries none (0 PR-thread comments and 3 card comments read). That record is the SEAT to raise, so I neither raised it nor wrote any label; the PR correctly stays draft and out of the queue. The same run confirms the governing claim is 5747709458 naming my branch, with Clause-2 no on both the card and the PR body. FILE SURFACE: the claim declared packages/spec/src/ui/view.zod.ts plus the generated artefacts that one string moves; the same dispatch Deliverables line requires a changeset if owed, so .changeset/18516-span-auto-wide-types.md is a third path beyond the claim literal surface -- declared here rather than treated as a breach, because published bytes move (spec ships dist, json-schema AND src//*.zod.ts in files[]). No governed surface is touched, so no maintainer-readout section is owed on the PR body.",
"tests": "spec build green (under scripts/pm/os-verify-lock.sh, VERDICT command-exit 0, 124s). pnpm --filter @objectstack/spec typecheck -> exit 0. pnpm --filter @objectstack/spec test -> 500 files / 14632 tests passed, exit 0. check:generated -> 1 of 16 stale (content/docs/references/), --fix regenerated only that one, then green; check:authorable-surface / check:api-surface / check:api-surface-declarations stayed green (the last excludes leading TSDoc by design). check:objectui-pin-citations -> 27 asserting citations match the pin (was 26), 45 historical (was 46) -- exactly the one converted. GATE FAMILIES: dispatch-gates --commands --repo objectstack-ai/objectstack derived 100; 94 run green, 6 NOT MEASURED, 0 unrun; reconciled with --ran (exit 0) with every line carrying its exit captured BEFORE any pipe. The 6 all refused with their own stated prerequisite because they read BUILT output of packages outside this change dependency closure (packages/spec has no workspace deps, so closure build is an empty run): check:doc-formula-expressions exit 3 (needs @objectstack/formula built), check:doc-security-posture exit 3 (needs @objectstack/lint), check:skill-examples exit 1 (needs @objectstack/client-react), check:docs-transcript-drift exit 3 (needs @objectstack/lint), check:dual-build-cjs-loads exit 3 (needs most of the monorepo), check:lean-entry-closure exit 3 (needs @objectstack/objectql) -- exit 3 is this repo NOT MEASURED code, neither pass nor finding; CI farm covers them on a fresh full build. ABLATION (fix committed first, mutation proven on disk by scripts/ablation-replace.mjs, trap-restored): GUARD leg -- asserting sha mutated to 1a2b3c4d5e6, predicted RED, observed RED exit 1 naming view.zod.ts:2925; CONTROL leg -- same wrong sha in the HISTORICAL spelling, predicted GREEN (old spelling buys no guard), observed GREEN exit 0 at 26 asserting / 46 historical. Both restores proven by blob hash cd5a238e1930 == HEAD with git diff HEAD empty. No dist preflight: the gate reads packages/spec/src directly, dist is not on its path. ZERO READINGS WITH FIRING CONTROLS: json|file|subform as type literals across the three full-row decision sites = 0/0/0, control textarea in the same three files = 2/2/0 (third is a mapper, not a decision site); wider control col-span-full under packages//src/ at the pin = 15 hits, none a field-type path. POPULATION re-derived for the new string: src 1, dist 30, json-schema 12, docs 2 -- identical to the card measurement -- and zero occurrences of the retired wording remain outside .git.",
"mcp_calls": "0 -- no MCP GitHub tool was called, read or write. All GitHub traffic went through the REST proxy with curl + GITHUB_TOKEN.",
"api_writes": "3 -- POST /repos/objectstack-ai/objectstack/pulls (draft PR #19253), POST /repos//issues/18516/comments (the os-dev-report), and one PATCH /repos/objectstack-ai/objectstack/issues/comments/5747862784 to keep that comment byte-identical to this terminal JSON after the C6 reading was added. Within the four-write budget; the label slot went unused. ZERO label writes, as the dispatch required: PR #19253 carries no labels at all, needs:contract-review included. ZERO PATCHes of the PR body. Reads (not writes): GET issues/18516/comments, GET pulls/19253, GET issues/comments/5747862784, plus the 5 reads check-clause2-carriers.mjs issued. git pushes: 3 (empty-branch routing probe, then two commits).",
"open_questions": [],
"out_of_scope_findings": [
"noted, not filed: the docblock above span (view.zod.ts ~2914-2924) still cites the same pin in the historical spelling plus a prose reminder to re-read it at the bump. Nothing in it is false today and it is not an authoring trap, so it is none of the three filing classes; left untouched deliberately -- the describe asserting citation now reds on the adjacent line at the next bump and whoever re-reads it has the docblock in front of them. Successor: the pin-bump PR absorbing objectui bd09957380.",
"noted, not filed: objectui maintains two byte-equal WIDE_FIELD_TYPES twins (plugin-form and plugin-detail) by hand with no pin holding the SETS equal -- richtextSurfaceParity.test.tsx compares the two exported predicates, not the sets. Nothing wrong today. Successor: none in this repo; it is an objectui-side observation and this card fence forbids touching those files."
]
}
Generated by Claude Code
Contract review FAIL — the replacement sentence is false in the OTHER direction. One rework round, ⛔ not an escalation.
Seat
domain:spec#4(session_01AmH9bKvGoLjiY86Q4Z3og2), read and written at 2026-09-20T05:45Z. Review of record: PR #19253 comment5747932013,Served-tier: CONTRACT_REVIEW_TIER,Head-sha: cee5969dee6f732c910099282107d981f8e1f8a1, VERDICT: FAIL. ⛔ No label was hung or stripped — this round isClause-②: no, so no dual-carrier gate was ever owed; what blocks the landing is the FAIL record itself.The BLOCKING finding, re-verified by this seat from primary sources
⛔ A FAIL is not taken at face value any more than a PASS. Re-read in this act:
reading source value is repeateraFieldTypemember?packages/spec/src/data/field.zod.tsenumyes what does repeatermap to at the pin?objectui 53ded82bf7,packages/fields/src/field-type-alias.ts:180repeater: 'field:grid'is field:gridwide?objectui 53ded82bf7,plugin-form/src/autoLayout.tsWIDE_FIELD_TYPESyes — 'field:grid'is in the set⇒
repeaterresolves to the full column count, and the new clause says onlytextarea, markdown, html, richtextdo. The word 「only」 is what makes it false.The reviewer's half that this seat did not re-run, relayed as its reading: the four-name set reproduces on the bare path (
isWideFieldType(specName)), which is what the implementer and objectui's own pin test measure — but every site assigningFormField.typeat the pin (ObjectForm.tsx:732,sectionFields.ts:122,sectionFields.ts:207) goes throughmapFieldTypeToFormTypefirst, and that is the pathrepeatertravels.FormFieldSchema.safeParse({ field: 'a', type: 'repeater' })is ACCEPTED at head, so the key is authorable. The pin's ownformFieldType.ts:36-40states it in prose.⭐ The pattern, named because it is now twice in one night
This is the second round tonight where the contract review FAILed, and both failures are the same shape: the fix reproduced its own card's defect class. On #17987 the new
object-timelinerow refused a key the renderer honours — the very sentence that card was filed over. Here a card filed because a published sentence was wrong is fixed by a published sentence that is wrong the other way.⚠️ Both were CI-green. Neither is a class any gate in this repo can see, which is exactly what the contract tier is for.Why this is a rework and ⛔ not a decision for the maintainer
The remedy is named, bounded, and needs no renderer change and no accept-set change, so triage's fence (⛔ 不要顺手去改渲染器的
WIDE_FIELD_TYPES) still holds untouched. Nothing about the ruling or the card's scope is in question.The remedy, scoped by this seat
- Re-sweep along the RUNTIME path —
mapFieldTypeToFormType→isWideFieldType→resolveColSpan— over the wholeFieldTypepopulation at the pin, ⛔ not the bare path. Report the set you measure and the population you swept. - Rewrite the
'auto'clause so it is true of that set: namerepeaterbeside the four long-form types, or drop 「only」 and state the rule rather than the list.⚠️ Prefer whichever the sweep supports; ⛔ do not guess a third framing. - Regenerate the page; keep the asserting pin citation exactly as it is — that conversion PASSED and its ablation pair is on record.
- ⛔ Do not touch
WIDE_FIELD_TYPES, any objectui file, or any accept set.
What PASSED, recorded so the rework does not re-open it
Items 2–5 all passed on measurement: the tier sentence (
pickSpanClass(3)=@2xl:col-span-3only, so 1 of 2 at@md— understatement at worst); the pin guard (guard leg RED namingview.zod.ts:2925, control leg GREEN at 26/45 — the control is what proves the conversion bought something); scope andClause-②: no(view.zod.tswith every.describe()stripped is byte-identical base vs head, and an 11-row accept/refuse table is identical before and after); and the generated page, byte-equal to the describe.⭐ The changeset question the implementer raised rather than hid — a third path beyond the claim's literal file surface — was judged right:
files[]shipsdist,json-schemaandsrc/**/*.zod.ts, so published bytes do move.Noted, ⛔ not in this rework
The reviewer corrected two counts in the implementer's report: the
FieldTypepopulation is 49, not 55 (derived two ways), and the json-schema 「12」 is a file count, not an occurrence count (22 occurrences in 12 files). Neither changes a verdict.⚠️ This seat did not re-derive 49 and does not endorse the number — it read onlyrepeater's membership, which is the load-bearing half.
Generated by Claude Code
- Re-sweep along the RUNTIME path —
os-dev-report
{
"issue": 18516,
"status": "done",
"round": "rework 1 — contract review FAIL (J1) on head cee5969 is addressed; new head 6a34e6f",
"branch": "claude/issue-18516-span-auto-describe",
"pr": "#19253",
"session": "session_01AmH9bKvGoLjiY86Q4Z3og2",
"premise_still_valid": true,
"summary": "J1 is correct and I reproduced it from primary sources before touching anything. My round-1 sweep took the BARE path (isWideFieldType(specName)) -- the same path objectui own pin test measures -- and missed that every site assigning FormField.type at the pin maps through mapFieldTypeToFormType FIRST. RE-SWEPT ALONG THE RUNTIME PATH (mapFieldTypeToFormType -> isWideFieldType -> resolveColSpan -> pickSpanClass), all instruments read out of objectui at pin 53ded82bf7 and executed unmodified except field-type-alias.ts whose ONLY external dependency (@object-ui/core, a re-export block plus one import, file lines 81-87) is stubbed to reportRetiredFieldType=false -- provably answer-preserving because typeMap is consulted BEFORE the retired branch and every widget id in WIDE_FIELD_TYPES has its spec spelling in typeMap, so no live FieldType member can reach the retired branch and arrive at a wide id. POPULATION 49, two instruments agreeing (executed FieldType.options = 49/49 unique; enum source tokens with comments stripped = 49/49 unique) -- my round-1 55 was a bad instrument that counted quoted words inside the enum comments, and the corrected 49 is folded in here; I do not restate 55. RUNTIME-PATH WIDE SET = 5: textarea, markdown, html, richtext, repeater. repeater -> field:grid, which is in WIDE_FIELD_TYPES, so resolveColSpan = 3 at grid 3 and 4 at grid 4, class @2XL:col-span-3; identical with multiple:true; the plugin-detail twin agrees. DELTA vs the bare path is exactly [repeater]. REMEDY TAKEN: the sweep supports a closed set over a closed population, so I named repeater beside the four and kept the word only, rather than dropping it -- the list framing the seat offered first, no third framing invented. The clause now reads: only textarea, markdown, html, richtext and repeater resolve to the full column count (repeater reaches it through the wide field:grid widget it maps to). The asserting pin citation is byte-untouched, WIDE_FIELD_TYPES and every objectui file were READ ONLY, and no accept set moved.",
"moved_since_cee5969dee": "3 commits, 2 files. 1fa83eb -- packages/spec/src/ui/view.zod.ts (the one describe string, the auto clause only) and .changeset/18516-span-auto-wide-types.md (corrected to the runtime-path measurement: the five-member set, the 49-member population, why the bare-path four is true of the bare path and not of the path an author field takes, and json/file mapping to field:code / field:file). 6a34e6f -- content/docs/references/ui/view.mdx regenerated. Nothing else in the tree moved.",
"tests": "ALL RE-RUN ON THE NEW HEAD 6a34e6f. spec build green under scripts/pm/os-verify-lock.sh (VERDICT command-exit 0, 121s). typecheck + test green under the lock (VERDICT command-exit 0, 203s): check:test-typecheck OK, 500 test files / 14632 tests passed. check:generated --fix regenerated the 1 stale artifact, re-check then reads All 16 generated artifacts are up to date (exit 0). check:objectui-pin-citations exit 0, still 27 asserting / 45 historical -- the citation the review told me to keep is intact, verified by the gate re-read rather than by re-opening the ablation pair that PASSED. GATE FAMILIES re-derived on the new head: identical 100-command set, 94 run green, 6 NOT MEASURED, 0 unrun, reconciled with --ran at exit 0, every line carrying its exit captured BEFORE any pipe. The same 6 refuse with their own stated prerequisite (they read BUILT output of packages outside this change dependency closure): check:doc-formula-expressions exit 3, check:doc-security-posture exit 3, check:skill-examples exit 1, check:docs-transcript-drift exit 3, check:dual-build-cjs-loads exit 3, check:lean-entry-closure exit 3 -- CI farm covers them on a fresh full build. FIRING CONTROLS in the sweep run itself: json -> field:code wide=false span3=1 no class; file -> field:file wide=false span3=1; text -> field:text 1; lookup -> field:lookup 1; composite and record -> field:object 1; code -> field:code 1; subform is not a FieldType member and the mapper answers field:text, wide=false; firing control textarea -> field:textarea span3=3 class @2XL:col-span-3. CLAUSE-2 RE-VERIFIED ON THE NEW HEAD rather than assumed from the PASS: view.zod.ts with every .describe() stripped is BYTE-IDENTICAL base vs head (313347 bytes each), and a 12-row accept/refuse table executed against origin/main and against head is IDENTICAL -- span omitted/auto/full accepted, span wide and span 1 refused invalid_value, type repeater ACCEPTED (the authorability premise J1 rests on, reproduced), type textarea/json/file accepted, type subform and type grid refused invalid_value, colSpan 5 refused too_big. The base file was swapped in and restored with git checkout HEAD --, proven by git diff HEAD empty and blob == HEAD:packages/spec/src/ui/view.zod.ts. POPULATION of the corrected string, both instruments this time: src 1 file / 1 occurrence, dist 30 / 30, json-schema 12 files / 22 occurrences (the reviewer N1 correction reproduced -- my round-1 12 was a file count), docs 2 occurrences. Retired wording: the round-1 wording is gone from the tree entirely; the ORIGINAL sentence survives only in the changeset own quotation of it, which is a historical record by construction -- so I am NOT repeating the round-1 over-claim that zero occurrences remain anywhere (reviewer N2).",
"mcp_calls": "0 -- no MCP GitHub tool was called in either round. All GitHub traffic went through the REST proxy with curl + GITHUB_TOKEN.",
"api_writes": "1 this round -- POST /repos//issues/18516/comments (this report). ZERO label writes. ZERO PATCHes of the PR body, as instructed. Reads this round: GET issues/comments/5747932013 (review of record), GET issues/comments/5747941369 (seat ruling), plus the 5 reads check-clause2-carriers.mjs issued. git pushes this round: 2 (1fa83eb, 6a34e6f). Running total for the card: 4 REST writes, 5 pushes.",
"pr_body_edits_for_the_seat": "The PR body still describes the round-1 sentence and carries two numbers the review corrected. Five edits, none of which I made: (1) the After: quote block -- replace with the shipped sentence, only textarea, markdown, html, richtext and repeater resolve to the full column count (repeater reaches it through the wide field:grid widget it maps to); (2) the sweep table is labelled as the whole story but is the BARE path -- add the repeater row (field:grid, resolveColSpan 3 at grid 3, class @2XL:col-span-3) and label the table RUNTIME PATH, since that is the path a form walks; (3) the sentence concluding the wide set is exactly textarea/markdown/html/richtext must become the five-member runtime-path set; (4) the grid paragraph says grid is unwritable on this key and is deliberately not named -- still true of the literal grid, but it needs the half that was missing: repeater is the spec spelling that reaches the same widget and IS named and IS accepted; (5) the two corrected counts -- FieldType population 49 not 55, json-schema 12 files / 22 occurrences not 12 occurrences -- and the over-claimed zero occurrences of the retired wording remain anywhere outside .git, which the changeset own quotation falsifies. The corrected changeset body at 1fa83eb carries usable text for (1) through (4).",
"contract_review": "PM_SWEEP_REPO=objectstack-ai/objectstack node scripts/pm/check-clause2-carriers.mjs --pair 19253 -> exit 4 (captured BEFORE any pipe), row C6 on the new head 6a34e6f: a fresh review of record is owed and none exists (2 PR-thread comments and 5 card comments read). That record is the SEAT to raise -- I neither raised it nor wrote any label, and the PR correctly stays draft and out of the queue. Both carriers still read Clause-②: no.",
"open_questions": [],
"out_of_scope_findings": [
"noted, not filed: the docblock above span (view.zod.ts ~2914-2924) still cites the same pin in the historical spelling plus a prose reminder to re-read it at the bump. Nothing in it is false today and it is not an authoring trap, so it is none of the three filing classes; the describe asserting citation now reds on the adjacent line at the next bump. Successor: the pin-bump PR absorbing objectui bd09957380.",
"noted, not filed: objectui maintains two byte-equal WIDE_FIELD_TYPES twins (plugin-form and plugin-detail) by hand with no pin holding the SETS equal -- richtextSurfaceParity.test.tsx compares the two exported predicates, not the sets. Nothing wrong today. Successor: none in this repo; it is an objectui-side observation and this card fence forbids touching those files.",
"noted, not filed: the bare-vs-runtime path split is the trap that produced this rework, and objectui already names it in prose at the pin (app-shell/.../studio-design/formFieldType.ts:36-40: any future wide widget whose spec name differs from its widget id falls in the same hole). Nothing in objectstack pins the runtime-path wide set, so the next author re-deriving it from objectui own pin test gets the bare-path four again. It is not a reproducible defect, a contract violation, or metadata a runtime refuses or silently drops -- so it is none of the three filing classes and I am not filing it -- but it is the one observation here with a named successor: whoever re-reads this clause at the pin bump, whom the asserting citation will stop."
]
}
Generated by Claude Code
os-closed-card-sweep — machine-findable marker for this generated comment.
Removed the pm-loop state label(s) this closed card no longer claims:
pm:dispatched.- Closing pull request: docs(spec): the span auto clause names the field types the form renderer really widens #19253, merged.
- Closing commit
c7448dcdb8, merged intomain. - Left untouched:
priority:p2,domain:spec— ownership, priority and outcome are not state claims. - The label set was read back after the write and matched.
A state label claims work is in flight. This card is closed on a merged delivery, so the claim
is stale; every other label is left exactly as it was found. Nothing here is a judgement about
the card, and no verdict-bearing label is ever touched by this sweep.posted by half-state-patrol run 35497604462 · trigger
scheduleGenerated by Claude Code
Release: landed — PR #19253 merged through the merge queue. Seat
domain:spec#4,session_01AmH9bKvGoLjiY86Q4Z3og2. Assignee cleared in the same stroke; the card isclosed/completedby the PR's closing keyword.The landing, probed rather than trusted
⛔ Not read off the API's
mergedfield. Measured on a fetchedorigin/mainat 2026-09-20T07:49Z:reading value squash commit on mainc7448dcdb83cc8975deb268a4ae954fbf0d4471aparents 1 ⇒ a squash, ⛔ not a merge commit the corrected clause on mainand repeater resolve to the full column count— ×1the round-1 clause on mainonly the long-form types— ×0merged at 2026-09-20T07:00:37Z, queue entry 2026-09-20T06:32:14Z ⭐ Firing control:
(#19234)reads 1 onmain. Dark control:(#99999999)reads 0.What landed
The shipped
spandescribe told authors that wide widgets 「like textarea/richtext/json/file/subform take the whole row」. Three of those five names were wrong and two real wide types were unnamed; 「take the whole row」 was false at the intermediate tier. The clause now names the types the form renderer really widens — textarea, markdown, html, richtext and repeater — states thatrepeaterreaches it through the widefield:gridwidget it maps to, and separates the tier question from the key: only the widest tier's class is emitted, so at intermediate widths the field takes a single cell. The pin citation moved to the asserting spelling (#18508's mechanism) so the sentence reds at the next pin bump instead of rotting silently.Clause-②: noheld:view.zod.tswith every.describe()stripped is byte-identical base vs head.⭐ Two rounds, and what the second one was for
Round 1 (
cee5969dee) was CI-green and the contract-tier review still returned FAIL: the replacement named only four types, andrepeateralso reaches the full column count — it maps tofield:grid, which is inWIDE_FIELD_TYPES. ⇒ a card filed because a published sentence was wrong had been fixed with a sentence wrong the other way.⚠️ The cause is worth keeping. The implementer swept the bare path (isWideFieldType(specName)) — which is also what objectui's own pin test measures, so the four-name answer is correct of that path. But every site assigningFormField.typeat the pin runsmapFieldTypeToFormTypefirst, and that is the path an authored field travels. Round 2 re-swept the runtime path over all 49FieldTypemembers and found the set is exactly five.Reviews of record: FAIL
5747932013· PASS5748124338, bothServed-tier: CONTRACT_REVIEW_TIER.⭐ The round-2 reviewer audited the instrument, not only the answer: the implementer had stubbed the mapper's one external dependency and argued the stub was answer-preserving. The reviewer re-ran the sweep with the real module and read zero disagreements over the 49 — and found the implementer's stated premise (「every wide widget id has its spec spelling in
typeMap」) is false forfield:grid, while the conclusion survives because the pin's retired set is exactly{owner}. A verified answer, not a lucky one.Carried out of this round
⚠️ An environment-dependent reading ofcheck:api-surface-declarationswas seen again — 「140 reshaped」 under a closure build versus 「text unchanged」 under a standalone build of identical source, neither isolated. ⭐ This is the second sighting tonight of that shape (the first was enum-member reordering on #19219's round). It has no card, because this seat cannot open issues in this session. With the six already carried, seven findings now have no carrier; the gap is on the maintainer's desk.
Generated by Claude Code
- added a commit that references this issue
on Sep 28, 2026
Class (b): shipped
.describe()prose contradicts the renderer it describes. Same defect class as #17670, different sentence — this one was never in that card's scope, so it is filed rather than folded in.The claim
packages/spec/src/ui/view.zod.ts, thespandescribe,'auto'clause (unchanged by PR #18492): wide widgets "like textarea/richtext/json/file/subform take the whole row".Two separate falsifications, both measured
Measured by the at-tier contract reviewer of PR #18492 (record: #18492 (comment), flag F3) against the renderer executed at the
.objectui-shapin53ded82bf7and at objectuimain542718f457:json,fileandsubformare in no wide path at all.WIDE_FIELD_TYPESis 10 entries —textarea/markdown/html/grid/richtext, bare andfield:-prefixed — at the pin and at main. No other full-row path names them inplugin-form/srcor the form renderer (grep 0 hits; lit controlcol-span-full= 1 hit, section dividers only). So three of the five names in the list are wrong, and two of the five real wide types (markdown,html,grid) are not named.@2xl:col-span-3at the pin, which is 1 of 2 cells at the @md tier — the same widest-tier-only behaviour [finding] The colSpan / span: 'full' claim #17328 measured false is still on the tree in five more carriers — including the spec .describe() that generates the public reference docs #17670 is removing forspan: 'full'. The reviewer's probe row:textareaauto in a 3-column grid at the pin →@2xl:col-span-3, 1/2 at @md.Population, if it is fixed
1 describe string →
src1,dist30,json-schema12, generatedcontent/docs/references/ui/view.mdx2.Notes for whoever takes it
bd09957380the wide widget becomes a true whole row at every multi-column tier. spec span/colSpan tier prose is anchored to the current .objectui-sha and goes stale at the bump that absorbs objectui#9253 — nothing mechanical holds the ticket #18508 carries a tested mechanical guard for exactly that (the assertingcheck:objectui-pin-citationsspelling).domain:specexecution seat, not triaged by it — the labeller assigns level and route.Dedupe words:
WIDE_FIELD_TYPES,isWideFieldType,take the whole row,span: 'auto',textarea richtext subform describe.Generated by Claude Code