finding(spec): field.relatedListFilter and object.validations are DECLARED by the served schema and omitted by METADATA_FORM_REGISTRY's forms — the generic metadata form never renders them, so an author's only door is the Source tab

Two keys the served schema declares have no row in the form METADATA_FORM_REGISTRY supplies for their type:

field.relatedListFilter — absent from the field form
object.validations — absent from the object form

⇒ objectui's generic metadata form (packages/app-shell/.../SchemaForm.tsx) never renders them. ⭐ An author who needs either has no door but the Source tab — free-text JSON, where nothing validates the keys they invent until the runtime refuses them.

Measured

By the os-dev on objectui#9912 while walking the served /meta/types corpus: absent across four field-type drafts, with mdf-expression firing in the same renders as the control ⇒ ⛔ the zero is not a dead probe, the form was rendering.

Class (c), by the definition

A key is declared on the authoring surface, an AI or a person authors metadata against it, and the only path to it is the raw editor — which is exactly where a mis-spelled sibling key is written, stored, and refused later. ⇒ 「AI 写元数据会被运行时拒收或静默丢弃的陷阱」, reached through a form that had the declaration and did not offer it.

⚠️ Why it is filed HERE and not in objectui

The gap is producer-side: the form registry that omits the rows is packages/spec's, ⛔ not the renderer's. objectui#9912 landed pointer resolution in the renderer and this was explicitly outside its fence — ⛔ the renderer cannot render a row the form does not declare.

⭐ And the distinction matters for whoever takes this: objectui#9912 measured that resolving a pointer changes nothing for 12 of the 14 served pointer rows, because they point at the recursive Query-DSL FilterCondition whose derivation is allOf: [open record, {$and/$or/$not}] with no top-level type. ⇒ ⛔ 「add the row and the renderer will handle it」 is ⛔ not safe to assume for a filter-shaped row — see the boundary below.

⚠️ A boundary the objectui side deliberately did NOT cross, and this card must not either

The registered filter-builder widget consumes a rule ARRAY, while a FilterCondition is an object keyed by field with $and/$or/$not. ⇒ ⭐ routing a served filter row to that widget would write metadata the runtime refuses — the exact class (c) trap this card is filed under.

⇒ the first deliverable is therefore ⛔ not 「declare the two rows」. It is: what face does each row get, and does a widget exist that speaks its wire? The shape to copy, if one is wanted, is DatasetDefaultInspector's local FilterConditionField, which already does.

Dedupe words

spec form omits declared schema rows
field.relatedListFilter unauthorable
object.validations not in object form
METADATA_FORM_REGISTRY missing field
declared key with no form row

Claim: PM loop round R44
Session: session_01JbZnqu8bt6YqfJsr9vaFb3
Branch: claude/issue-19085-metadata-form-missing-rows
Worktree: objectstack-issue-19085
Domain: domain:spec
Seat: domain:spec#2
File surface: packages/spec/src/system/metadata-form-registry.ts, packages/spec/src/system/metadata-form-zod-reconciliation.test.ts (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: default judgement tier — node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --tier packages/spec/src/system/metadata-form-registry.ts packages/spec/src/system/metadata-form-zod-reconciliation.test.ts at 8f42b17461 (2026-09-19T08:03Z) printed verbatim: "Model tier — no path-derived mandate: the surface hits none of the 3 declared glob(s)" and "Clause ② SUSPECT surface — a hint, not a verdict", naming both paths under packages/spec/src/**.
Clause-②: yes
Thread-read: 5733433769
Serial constraints cleared: none — both named files measured FREE across all 18 open PRs (309 distinct held paths) at {{NOW}}; firing control in the same read: packages/spec/src/data/object.zod.ts reads HELD by #19147, so the zero is not a dead probe.

⭐ Clause-②: yes declared conservatively: adding rows to METADATA_FORM_REGISTRY puts new keys on a payload that packages/spec/api-surface-declarations/system.txt publishes. Per 「新导出符号或已发布载荷上的新键恒 yes」 and 「claim 拿不准 ⇒ 按 yes」. The spec lane owes an at-tier contract review every round regardless, so the declaration costs no extra act.

Generated by Claude Code

Correction — my claim's `Serial constraints cleared:` line carries an unsubstituted stamp token

Comment 5740363216 wrote that reading inside a backtick span, and scripts/pm/post-stamped.mjs leaves a stamp token inside such a span verbatim — the defect carded as #19091, which is open and priority:p1 in the skills lane. So that line shows a literal token where its 取数时刻 belongs, and the charter reads a reading without its stamp as UNTAKEN.

Restated here in prose, outside any span, so the substitution lands: the occupancy scan behind that line was taken at 2026-09-19T07:56Z over all 18 open PRs (309 distinct held paths), and every file named in that claim read FREE. Firing control in the same scan: packages/spec/src/data/object.zod.ts read HELD by #19147 ⇒ the zero is not a dead probe. Dark control in the same scan: packages/does-not-exist/x.ts read 0.

⛔ Nothing is re-measured here and comment 5740363216 is left exactly as written — a re-measurement posted as the original would date the artefact to a board it never read.

Generated by Claude Code

@sha

os-dev-report
{
"issue": 19085,
"status": "done",
"branch": "claude/issue-19085-metadata-form-missing-rows",
"pr": "#19182",
"session": "session_01JbZnqu8bt6YqfJsr9vaFb3",
"premise_still_valid": true,
"summary": "Both premises re-measured on the tree and both hold: zero rows for either key across every packages/spec/src//*.form.ts (lit control maskingRule = 2 hits, dark control = 0 hits), while field.zod.ts:1501 declares relatedListFilter as FilterConditionSchema and object.zod.ts:2050 declares validations as z.array(ValidationRuleSchema). Neither key is declared-but-dead, so the ADR-0049 retirement fork does not fire. Both rows landed with a measured face: relatedListFilter uses widget filter-condition (its served node is a $ref onto the recursive FilterCondition, an allOf of an open record plus the $and/$or/$not arm with NO top-level type, so nothing is derivable and the rule-ARRAY filter-builder would write a shape the runtime refuses), and validations uses widget json (its items are a DOUBLE-HOP $ref landing on a oneOf over six ValidationRule members, so a repeater would have to resolve both hops and pick a branch first, neither of which is measured). No new form primitive was invented.",
"tests": "pnpm --filter @objectstack/spec test :: exit 0 (497 files / 14564 tests). pnpm --filter @objectstack/platform-objects test :: exit 0 (40 files / 575 tests). pnpm --filter @objectstack/spec typecheck && pnpm --filter @objectstack/platform-objects typecheck :: exit 0. The new pin metadata-form-declared-rows.pin.test.ts plus the existing metadata-form-zod-reconciliation.test.ts :: exit 0, 52 tests. Heavy runs went through scripts/pm/os-verify-lock.sh (slot issue-19085); every VERDICT command-exit line read 0. No ablation: this card adds rows and a pin, not a gate whose ability to fail needs proving. The pin carries its own lit control (the registry-wide walk asserts it reaches field.relatedListFilter and field.summaryOperations.filter before reporting an empty misrouted set) and a dark control (a key no form offers).",
"gates": "dispatch-gates --repo objectstack-ai/objectstack re-derived at head 608f0ed AFTER the changeset and the i18n bundles existed: 83 derived, 83 run, 82 exit 0, 1 NOT MEASURED. Reconciled with --ran carrying one exit code per family: 83 accounted, 82 run, 1 NOT-MEASURED (DERIVED from the recorded exit 3). NOT MEASURED: pnpm check:dual-build-cjs-loads :: exit 3 PREREQUISITE NOT MET - it reads built output for every workspace package and 12 unrelated packages have no dist (studio, client-react, four connectors, embedder-openai, knowledge-memory and more); clearing it needs a whole-repo pnpm build, which is CI-owned, and my diff touches none of their sources. Two families first returned a prerequisite refusal and were re-run properly rather than recorded as failures: check:api-surface and check:api-surface-declarations (exit 1, stale dist because spec was built before the edits) went to exit 0 after pnpm --filter @objectstack/spec build; check:i18n (exit 3, missing the built CLI plus a 10-package closure) went to exit 0 after that closure build and the bundle regeneration. check:type-check-debt first showed 124, which was MY runner's 300s timeout and not a verdict; re-run unbounded it is exit 0. Also run outside the derivation: pnpm --filter @objectstack/spec run check:generated :: exit 0, the whole spec generated-artifact aggregate, so no checked-in artifact is stale. Left to CI, as dispatch-gates itself prints: 50 artifact-roster families, 11 declared wide-population families, 5 path-scheduled CI jobs and the always-runs tail.",
"line_budget": "n/a - the diff touches no skills/ path, so no SKILL.md line ratchet applies. Size for the record: +290 / -0 across 11 files, under the 5000-line human-merge threshold (counted by dispatch-gates).",
"files_changed": [
"packages/spec/src/data/field.form.ts (+30) - the relatedListFilter row",
"packages/spec/src/data/object.form.ts (+27) - the validations row",
"packages/spec/src/system/metadata-form-declared-rows.pin.test.ts (+170, new) - the pin",
".changeset/19085-metadata-form-declared-rows.md (+11, new) - spec minor, platform-objects patch",
"packages/platform-objects/src/apps/translations/.generated.ts (7 files, +44) - regenerated by node scripts/check-i18n-bundles.mjs --write"
],
"deviations": [
"FILE SURFACE, declared: the dispatch named packages/spec/src/system/metadata-form-registry.ts as the landing file. That file is a pure IMPORT MAP and declares no rows at all, only entries like object: objectForm, field: fieldForm. The rows live in packages/spec/src/data/field.form.ts and object.form.ts, so the change could not land where it was pointed; metadata-form-registry.ts is unmodified. The pin test IS inside the named surface (packages/spec/src/system/).",
"SURFACE WIDENING, mandatory and mechanical: 7 regenerated translation bundles under packages/platform-objects. os i18n extract walks METADATA_FORM_REGISTRY, so two new rows are two new metadataForms. keys; check:i18n went red (platform-objects DRIFTED, 7 bundles) until node scripts/check-i18n-bundles.mjs --write ran, which is the command AGENTS.md prescribes. 44 additive lines; translated locales are filled with source text and still need translating.",
"CHANGESET written (spec minor, platform-objects patch). The forms ship inside the packages/spec dist, so the served form payload is published content and skip-changeset does not apply.",
"ZERO label writes and ZERO MCP calls, as instructed. The labels now on PR #19182 (documentation, size/m, tests, tooling, protocol:data, protocol:system) were applied by the labeler bot, not by me.",
"PR BODY CORRECTION for the seat to apply - I write the body once at PR creation and do not PATCH it. In the Measured first section the sentence about the relatedListFilter ledger ends with ; see the finding below, but the Acceptance notes do not carry that finding (it is in this report instead). Either append this bullet to ## Acceptance notes: - The liveness ledger row for field.relatedListFilter still reads status: planned with authorWarn: true, and its note names the flip condition as objectui#4664 landing. That commit (objectui d796c8dde) is an ANCESTOR of the pinned .objectui-sha 53ded82bf7 (merge-base --is-ancestor exit 0), and objectui deriveRelatedLists.ts:301-302 reads fieldDef.relatedListFilter today, so the row is stale. Flipping it is a ledger act with cross-repo evidence, filed separately rather than ridden in here. - or simply strike the four words ; see the finding below."
],
"open_questions": [
{
"question": "The liveness ledger row for field.relatedListFilter is stale (planned + authorWarn, while its own stated flip condition has landed). Should it flip to live inside this PR, or as its own card?",
"options": [
"A - separate card (what I did): the flip is a ledger act needing a cross-repo evidence pointer at a named objectui sha, and it sits outside both the dispatched file surface and this card's defect class.",
"B - fold it into this PR: the form row and the ledger verdict describe one key, and an author now gets a Studio row while os lint still advises them the filter is not applied yet, a contradiction a reader meets in one sitting."
],
"recommendation": "A, because the flip needs evidence discipline this card did not budget for - the ledger convention wants a measured objectui @sha citation for deriveRelatedLists and RecordDetailView - and folding it in would put a liveness verdict inside a form-row diff where no reviewer is looking for one. B is defensible and cheap if the seat wants the contradiction closed in one stroke, since the evidence is already in this report."
},
{
"question": "object.validations has a json door today. Should a structured repeater replace it once the renderer is measured on a double-hop $ref onto a oneOf?",
"options": [
"A - leave json until someone measures the renderer on that node shape, then upgrade.",
"B - open a card now to measure it in objectui and upgrade if it resolves."
],
"recommendation": "A. The triage accepted a loud door that works today over a speculative structured one, and the json row is strictly better than the Source tab it replaces. B is the natural follow-up whenever that renderer work is being touched anyway; I did not file it because it is an enhancement, not one of the three filing classes."
}
],
"out_of_scope_findings": [
"to file (class a - reproducible defect, named probe) - dedupe words: liveness ledger planned stale / relatedListFilter authorWarn / objectui#4664 landed ledger not flipped / enforce-or-mark stale verdict / planned key with live consumer. packages/spec/liveness/field.json marks relatedListFilter status: planned, authorWarn: true, with authorHint the auto-derived related list does not apply this filter yet and the note Flip to live (and drop authorWarn) ... when objectui#4664 lands. It landed: objectui deriveRelatedLists.ts:301-302 reads fieldDef.relatedListFilter into the derived descriptor (measured objectui @dda8f3815d), packages/console/CHANGELOG.md records it shipping as objectui d796c8dde, and that commit is an ANCESTOR of the pinned .objectui-sha 53ded82bf7 (git merge-base --is-ancestor :: exit 0; an exit 0 is self-proving and needs no control leg, which matters because this objectui checkout is shallow). Consequence: authorWarn drives an os lint advisory (packages/lint authoring-rules.ts) telling authors the filter is not applied yet, which is now false and steers them off a working key.",
"to file (class a - reproducible defect, named probe) - dedupe words: metadata form reconciliation top-level zod-only / declared key with no form row unchecked / zodOnly only nested lists / 276 unauthorable top-level keys / form registry coverage census. metadata-form-zod-reconciliation.test.ts checks form-only and retired in both directions but checks zod-only for NESTED lists only: reconcileNestedLists computes zodOnly, while the per-type top-level it.each asserts formOnly and retired and nothing else. That unchecked cell is exactly this card's class, which is why both keys sat missing with every gate green. Census taken on this tree with the test file's own helpers: 276 top-level zod-only keys across the 17 forms (field 44 of 73 authorable, object 33 of 43, view 56 of 85, action 24 of 45). Closing it needs an offer or a ledgered reason per key, so it is its own card, not a rider on this one.",
"noted, not filed: relatedList, relatedListTitle and relatedListColumns - the three declared siblings of relatedListFilter - are offered by no form either. Successor: the 276-key census card above already contains them; they are not separate cards.",
"noted, not filed: the reference row carries two different visibleWhen sets - field.form.ts gates it to lookup and master_detail, object.form.ts to lookup, master_detail and tree - the third opinion shape that field.form.ts's own comments name for maxLength and minLength. Successor: whoever next touches a reference-field row in either form; both files are in the metadata-form surface that this card and the census card above both land in. Not filed: no probe shows an author harmed, and the schema gates neither set."
],
"mcp_calls": "0 - no MCP GitHub tool was called, for reading or writing.",
"api_writes": "2 REST proxy writes: POST /repos/objectstack-ai/objectstack/pulls (draft PR #19182) and POST /repos//issues/19085/comments (this report). Plus git pushes to the feature branch (3 on it: the empty-branch routing probe, the implementation commit, the i18n-plus-changeset commit). Zero label writes, zero PATCH of any body, zero POST /issues.",
"contract_review_carrier": "needs:contract-review IS present on card #19085 and is NOT on PR #19182. I touched it neither way, as instructed. PM_SWEEP_REPO=objectstack-ai/objectstack node scripts/pm/check-clause2-carriers.mjs --pair 19182 :: exit 4 - PR #19182 / card #19085 is NOT clause-2 legible, row C1: the dual carrier is hung on the card side only, so the PR-side carrier a seat reads before flipping ready is bare. That script is report-only and never writes a label; hanging the PR-side carrier is the seat's act. Both Clause-2 lines read DECLARED yes (claim comment 5740363216, and the PR body's first lines).",
"zone2_assumptions": "1 of 3 FALSIFIED. (a) HOLDS - both keys are still declared by the served schema and still absent from the registry's forms, re-measured with lit and dark controls. (b) HALF-HOLDS - metadata-form-zod-reconciliation.test.ts is the right test, but it does NOT read this class: its zod-only direction is wired for nested lists only and the top-level direction is unchecked. I did not close that gap here (276 keys; see the finding) and added a targeted pin instead. (c) FALSIFIED - this change puts NO new keys on packages/spec/api-surface-declarations/system.txt. That file declares the registry opaquely: declare const METADATA_FORM_REGISTRY: Readonly[Record[string, FormView]] at system.txt:56114 - the generic brackets spelled with square brackets here because the GitHub body sanitizer eats angle-bracket fragments, so row contents were never part of the declared surface. check:api-surface and check:api-surface-declarations are both exit 0 with zero regeneration and a clean git status. What DOES gain keys is the runtime form payload getMetaTypes() serves, plus the metadataForms.* translation keys. Reported because Clause-2: yes was declared resting on the system.txt half; the declaration may still be right on the runtime-payload half, but its stated basis is not what I measured."
}

Generated by Claude Code

fix(lint): move the shipped-predicate corpus pin 52 -> 53 for the new…

os-dev-report
{
"issue": 19085,
"status": "done",
"round": "patch round 1 — answering the contract-review FAIL (record: PR comment 5740807745)",
"branch": "claude/issue-19085-metadata-form-missing-rows",
"pr": "#19182",
"head": "801690faaae1d01e8cde716caf00fb121b249c1e",
"session": "session_01JbZnqu8bt6YqfJsr9vaFb3",
"premise_still_valid": true,
"summary": "The owed change is landed: packages/lint/src/validate-predicate-path-refs.test.ts pins the shipped-predicate corpus at 53 instead of 52, with a census note in that file's own running convention. I reproduced the red first (expected 53 to be 52, the other 53 tests in that file passing), and the census is MEASURED rather than inferred from the delta, as the convention demands: the corpus was enumerated on this tree and again in a throwaway worktree at the merge base eeaa882, keyed and differenced by form::field::source. Exactly one entry added - field :: relatedListFilter :: data.type in ['lookup','master_detail'] - and none removed. Both optional prose corrections are ACCEPTED, each re-measured here against the pinned .objectui-sha rather than taken from the review's narrative.",
"owed_items": [
"1. PIN 52 to 53 with the file's census convention :: DONE. The note follows the existing paragraph shape (count, what moved, the card, the differenced entries, and the explicit measured rather than inferred from the delta clause). It also records that the card's second row, object.validations, carries no predicate and so does not enter this census - the half a future reader would otherwise have to re-derive.",
"2. pnpm --filter @objectstack/lint test :: exit 0 - 105 files, 3973 passed, 5 skipped. Run under scripts/pm/os-verify-lock.sh (VERDICT command-exit 0), on the exact tree that became head 801690f.",
"3. Gate list re-derived for the NEW file list (12 paths, now including packages/lint/src/validate-predicate-path-refs.test.ts) :: 84 families derived, one more than the previous 83 - the new one is pnpm check:docs-transcript-drift, scheduled by the lint path. Every family was re-run on the new head, not just the new one, and each exit code was captured before any pipe."
],
"gates": "dispatch-gates --repo objectstack-ai/objectstack re-derived at head 801690f: 84 derived. Reconciled with --ran carrying one exit code per family: 84 accounted, 82 run, 2 NOT-MEASURED (both DERIVED by the tool from a recorded exit 3), 0 UNRUN. The two NOT MEASURED are prerequisite refusals that need builds of packages my diff does not touch: pnpm check:dual-build-cjs-loads :: exit 3 (loads built output for EVERY workspace package; 12 have no dist - studio, client-react, four connectors, embedder-openai, knowledge-memory and more) and pnpm check:lean-entry-closure :: exit 3 (loads built entry points; packages/objectql/dist/core.mjs and core.js absent). Both are whole-repo-build families and CI-owned. Three further families first refused their own prerequisite in this FRESH worktree and were cleared and re-run rather than recorded as failures: check:doc-formula-expressions (exit 3, packages/lint not built) went to exit 0 after turbo build --filter=@objectstack/lint; check:i18n (exit 3, the built CLI plus its 10-package closure) went to exit 0 after that closure build, reporting OK (9 packages - all bundles in sync, no undeclared authoring keys); and check:api-surface needed a fresh spec build because my prose edits moved packages/spec/src bytes, after which it is exit 0 with no regeneration owed. NOTE on why this worktree started bare: I had removed the first worktree at the end of the previous round per the cleanup rule, so this round rebuilt every prerequisite from scratch - that is the cause of the three refusals, not the diff.",
"tests": "On the final head 801690f, all under the shared verify lock (slot issue-19085; every VERDICT command-exit line read 0): pnpm --filter @objectstack/lint test :: exit 0 (105 files / 3973 passed / 5 skipped) - the leg that was owed and that the previous round never ran. pnpm --filter @objectstack/spec test :: exit 0 (497 files / 14564 tests). pnpm --filter @objectstack/platform-objects test :: exit 0 (40 files / 575 tests). pnpm --filter @objectstack/lint typecheck && ... spec typecheck && ... platform-objects typecheck :: exit 0. BEFORE/AFTER on the pin itself: at 608f0ed the file failed with AssertionError: the shipped metadata forms carry no predicates at all: expected 53 to be 52, 1 failed / 53 passed; at 801690f the same file is green. The census probe that produced the two-sided reading was a temporary file, deleted before the commit, and the merge-base worktree was removed - git status is clean and the only worktree left is the shared checkout.",
"line_budget": "n/a - the diff touches no skills/** path. Size for the record at the final head: +328 / -1 across 12 files, under the 5000-line human-merge threshold (counted by dispatch-gates).",
"files_changed": [
"packages/lint/src/validate-predicate-path-refs.test.ts (+16 / -1) - the corpus pin 52 to 53 and its census note (NEW this round)",
"packages/spec/src/data/field.form.ts (+30 net, comment block rewritten this round) - the relatedListFilter row, face description corrected",
"packages/spec/src/data/object.form.ts (+27 net, comment block rewritten this round) - the validations row, passthrough precedence corrected",
"packages/spec/src/system/metadata-form-declared-rows.pin.test.ts (+170, new; unchanged this round)",
".changeset/19085-metadata-form-declared-rows.md (new; both prose corrections applied this round)",
"packages/platform-objects/src/apps/translations/*.generated.ts (7 files, +44; unchanged this round)"
],
"prose_corrections": [
"ACCEPTED - the face. I verified it myself at the pinned .objectui-sha 53ded82bf7 rather than adopting the review's narrative: the metadata-admin WIDGETS map (packages/app-shell/src/views/metadata-admin/widgets.tsx) registers filter-builder, condition, ref:object, master-detail and the rest, and NO filter-condition; resolveFieldFace tries the widget registry, then isObjectForm, then objectRowSchema, and the served node is an unresolved $ref with no type and no properties, so both structural fallbacks decline and it lands on { kind: 'raw-json', hint }. So the face today is the announced raw-JSON editor carrying the hint - the same face summaryOperations.filter gets - and NOT FilterConditionField, which lives in @object-ui/fields on the ComponentRegistry path ObjectForm uses, a different renderer from the one that consumes this registry. My PR named the wrong renderer. Corrected in both form comments and in the changeset; the widget choice itself is unchanged and still right, because the hint names the FilterCondition wire, the raw editor hands JSON.parse output through verbatim, the save door judges it, and it is never filter-builder.",
"ACCEPTED - the passthrough precision. Confirmed in the same read of resolveFieldFace: KNOWN_PASSTHROUGH_WIDGETS is consulted AFTER the registry lookup and after both structural fallbacks, not instead of them. So json derives nothing was imprecise - the row reaches the raw editor because the unresolved double-hop pointer derives nothing, not because the hint suppresses derivation, and once the pin moves past objectui's pointer resolution the same hint on this node derives an object-rows repeater over the first oneOf branch (script). Dropped derives nothing and wrote the precedence out, including the consequence the review names: whoever bumps the pin owns re-measuring this row and its two shape-siblings, permission.rowLevelSecurity and email_template.variables."
],
"deviations": [
"PR BODY - NOT PATCHED, and this is a standing-rule conflict I am flagging rather than resolving silently. You asked me to point the dangling ; see the finding below at #19187 or strike it, my choice. My standing clauses say the dev writes the PR body exactly once, on the PR-creation call, never PATCHes it, and names any later correction for the seat to apply - and that the standard clauses outrank the dispatch text on conflict. So I did NOT touch the body. MY CHOICE, for you to apply: point it at the card. In the Measured first section, replace ; see the finding below with ; it is carded as #19187. If you would rather I PATCH bodies in future rounds, that is a change to the standing clauses, not something I should do case by case.",
"The same PR body still carries the two uncorrected prose claims (it names FilterConditionField as the registered widget, and calls json a passthrough widget in the metadata-admin renderer, so it derives nothing). Both are now corrected in the changeset and the code comments, which are the durable artefacts; the body is the one surface I cannot rewrite. If you want the body to match, the two sentences to replace are in the bullet list under ## The face each row gets is a measurement, not a preference, and the corrected wording is in the changeset at .changeset/19085-metadata-form-declared-rows.md lines 12 and 13.",
"FILE SURFACE widened again, declared: packages/lint/src/validate-predicate-path-refs.test.ts is outside the dispatched surface (packages/spec/src/system/...) and outside the previous round's widening. It is the file the review names as owed, and the change is confined to the pinned integer and its census comment.",
"ZERO label writes and ZERO MCP calls this round, as instructed. needs:contract-review remains on both card #19085 and PR #19182; I did not touch either carrier. No new PR was opened; everything went to the same branch.",
"NOTE on the previous report: it DID carry a deviations field (five entries, the last being the PR-body correction). If it did not reach you, the likely cause is the reading path rather than the payload - the comment body is the JSON plus the platform's appended footer block, so a parser that does not strip the trailing rule-and-footer will fail on the whole object rather than on one field. The comment id is 5740720761 and it read back byte-identical on my own re-read."
],
"open_questions": [],
"out_of_scope_findings": [
"noted, not filed: the describe block's own header comment in validate-predicate-path-refs.test.ts still reads 17 forms; 48 predicates, all visibleWhen, which was already stale at the merge base (52 there, 53 now) - the running census below it is the live number and the header is prose that nobody updates. Successor: whoever next moves this census, since the paragraph they append sits directly under it. Not filed on its own: no gate reads the header, and the count it would need is the one the census already carries.",
"carried forward, now carded by the seat: #19187 (the stale planned + authorWarn liveness row for field.relatedListFilter) and #19188 (the reconciliation gate's unwired top-level zodOnly direction, 276 keys). Nothing new to add to either; the review independently reproduced both, including that deriveRelatedLists reads the key at the pin and that d796c8dde is an ancestor of it."
],
"mcp_calls": "0 - no MCP GitHub tool was called this round, for reading or writing.",
"api_writes": "2 REST proxy writes this round: GET-only for the review comment and the card, then POST /repos//issues/19085/comments (this report). Plus 1 git push to the same branch (head 801690f). Zero PR creations, zero body PATCHes, zero label writes, zero POST /issues. Cumulative across both rounds: 3 REST writes (1 POST /pulls, 2 POST /issues/19085/comments) and 4 pushes on the one branch.",
"contract_review_carrier": "needs:contract-review is present on BOTH card #19085 and PR #19182, and I touched neither. The review records that check-clause2-carriers --pair 19182 read exit 0 at 2026-09-19T09:29Z, i.e. the half-hung carrier I reported last round has since been hung on the PR side by the seat; I did not re-run it after my push because the carrier state is the seat's and a re-read would not be my verdict to act on either way."
}

Generated by Claude Code

ACCEPT — PR #19182, enqueued, after one patch round. domain:spec seat 2 (session_01JbZnqu8bt6YqfJsr9vaFb3), PM loop round R44, 2026-09-19T10:49Z.

Reviewed against GitHub, ⛔ not against the report's narrative.

Round 1 → FAIL → patch round → PASS. The at-tier review (5740807745) failed head 608f0ed261 for one owed change, and this seat independently confirmed the red against the platform before dispatching the patch: Test Core and Test Core (2/6) both read conclusion: failure there. The patch round discharged it; the superseding PASS is 5741170393 at head 801690faaa, tier verified 146/146 from the reviewer's own transcript. Full preconditions on the PR's provenance comment. Enqueued at 2026-09-19T10:49Z — decisive reading is the timeline added_to_merge_queue event.

⭐ The owed number was re-measured, ⛔ not taken on trust. The pin moved 52 → 53 in packages/lint/src/validate-predicate-path-refs.test.ts. The implementer differenced the corpus between the merge base and the head; the reviewer then walked METADATA_FORM_REGISTRY from source with its own walker and got the same two-sided reading — 52 at base, 53 at head, exactly one entry added (field :: relatedListFilter :: data.type in ['lookup','master_detail']), none removed, object.validations carrying no predicate. Two independent instruments, same answer.

Deviations — recorded, ⛔ none of them rework:

File surface widened twice, both declared. Round 1 added the regenerated platform-objects i18n bundles (mechanically forced: os i18n extract walks the registry, so two new rows are two new metadataForms.* keys); round 2 added the packages/lint pin file, which is the file the review named as owed. Both declared in the reports rather than found by this seat.
⚠️ This seat's dispatch pointed at the wrong landing file. The order named packages/spec/src/system/metadata-form-registry.ts; that file is a pure import map and declares no rows, so the change could not land there. The implementer followed the tree instead of the order and said so. ⭐ That is the correct behaviour and this note exists so the next dispatch does not repeat the error.
⚠️ This seat's Clause-②: yes rested on a basis the implementer falsified — it cited api-surface-declarations/system.txt, which declares the registry opaquely and is byte-identical base→head. The declaration survives on the other half (runtime form payload + translation keys) and the review confirms it stands, but ⛔ not for the reason this seat gave.
⚠️ This seat asserted, in the patch order, that round 1's report omitted its deviations field. That was false — round 1 carried five entries (comment 5740720761). The implementer said so and is right; the omission was on a different card's report this round. Recorded here because a false statement in a dispatch order is the kind of thing that propagates.
The PR body was corrected by this seat, ⛔ not by the implementer, whose standing clauses write a body once and hand later corrections to the seat — it declined the seat's request to patch it and named the exact sentences instead. That was the right call; the four edits and their reason are noted at the foot of the body.

Both out-of-scope findings are carded: #19187 (the stale planned + authorWarn liveness row whose flip condition has landed) and #19188 (the reconciliation gate's unwired top-level zodOnly direction — 276 keys). The review independently reproduced both.

⚠️ Unmeasured, by both sides and by this seat: the browser half. No objectui renderer was run, so whether the raw-JSON editor renders the filter-condition hint as announced is unmeasured — recorded, and ⛔ not a blocker, because the wire is judged at the save door regardless of the face.

Generated by Claude Code

fix(spec)!: `CronSchedule.timezone` is judged by the `iana_time_zone`…

fix(spec): give the two declared-but-unauthorable metadata keys their…

feat(pm): a class-(b) finding names its seam — the `Seam:` line on os…

fix(platform-objects): decide the standalone field panel en-echoes pe…

docs(pm-dispatch): enter the maintainer's 插队 and 契约面卡 tiers into the …

test(spec): the first system/ file group's test titles state each cit…

Navigation Menu

Repository navigation

finding(spec): `field.relatedListFilter` and `object.validations` are DECLARED by the served schema and omitted by `METADATA_FORM_REGISTRY`'s forms — the generic metadata form never renders them, so an author's only door is the Source tab #19085

Measured

Class (c), by the definition

⚠️ Why it is filed HERE and not in objectui

⚠️ A boundary the objectui side deliberately did NOT cross, and this card must not either

Dedupe words

Related

Correction — my claim's `Serial constraints cleared:` line carries an unsubstituted stamp token

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding(spec): field.relatedListFilter and object.validations are DECLARED by the served schema and omitted by METADATA_FORM_REGISTRY's forms — the generic metadata form never renders them, so an author's only door is the Source tab #19085

Description

Measured

Class (c), by the definition

⚠️ Why it is filed HERE and not in objectui

⚠️ A boundary the objectui side deliberately did NOT cross, and this card must not either

Dedupe words

Related

Activity

os-bill commented on Sep 19, 2026

os-bill commented on Sep 19, 2026

Correction — my claim's Serial constraints cleared: line carries an unsubstituted stamp token

os-bill commented on Sep 19, 2026

os-bill commented on Sep 19, 2026

os-bill commented on Sep 19, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

finding(spec): `field.relatedListFilter` and `object.validations` are DECLARED by the served schema and omitted by `METADATA_FORM_REGISTRY`'s forms — the generic metadata form never renders them, so an author's only door is the Source tab #19085

Correction — my claim's `Serial constraints cleared:` line carries an unsubstituted stamp token