docs(skills): mark the two non-canonical row-predicate spellings as deprecated

[zhuangjianguo]

Fixes#5759

Prose-only annotation of the three-way row-predicate sentence in the published skills guide skills/objectui/guides/schema-expressions.md (List-view/CEL tier). It presented record.status, bare status and data.status as peers, labelling record.* merely "canonical" — a word that reads as a style preference — while since Phase 1 (#5330 / PR #5737) the other two emit a one-time dev-console deprecation warning and retire once a stored-metadata survey sizes the window.

What changed

One sentence, at :307. record.* is now stated as the canon; bare status and data.status are stated as deprecated here (still bound, warned once in dev, retiring after the survey), pointing at packages/core/src/evaluator/rowPredicateCanon.ts; and the card's sharper point is carried: data.* is the trap — the server's authoring oracle accepts it silently and it then binds nothing at runtime, a constant false rather than an error.

"Deprecated here" is deliberate, not hedging: rowPredicateCanon.ts scopes the deprecation to the runtime record layer, because data is the canonical root of a metadata-editing form (ADR-0089 D3). An unqualified sentence in this corpus would contradict that.

No example was rewritten — the nearby data-model-tier and conditional-formatting examples are already record.*.

State verified on this branch, not taken from the card

The card was filed 2026-08-23; the words had to describe what is measurable now. Read at base aca70f6:

• listConditional.ts:273 still binds all three — { ...(opts.scope ?? {}), ...rowObj, data: rowObj, record: rowObj }. No spelling has been retired or turned into a refusal, so "still bound" is accurate and "retiring after a stored-metadata survey" is the correct tense.
• listConditional.ts:317 calls warnNonCanonicalRowSpelling(predicateText, rowObj, !opts.rowless, opts.label) on the CEL path, so both non-canonical spellings do warn on this tier, and data names the row here (not rowless), which is the condition the detector needs.
• rowPredicateCanon.ts still reads Phase 1: "the binding is UNCHANGED and every spelling still resolves", warn-once, "removable only after a stored-metadata survey".

Phase 2 (#5741) has not advanced past "warned".

Three-engine boundary held

Only the List-view/CEL row-predicate tier changed. The ${…}/On-suffix tier earlier in the file (SafeExpressionParser, where data.* is correct) and the flow tier (isFieldVisibleWhen, previews/screen-spec.ts) are untouched. The one-time-warning sentence at :316-317 belongs to legacy-syntax routing, a different mechanism from these spellings, and is untouched — the diff hunk stops above it.

Line-count readings (2026-08-21 whole-package ruling)

PM-set budget for this card: +4 lines max net on the file.

reading	before	after	net
skills/objectui/guides/schema-expressions.md (whole file)	621	625	+4
published bundle — all 18 .md under skills/	5682	5686	+4

git diff --stat: 1 file changed, 7 insertions(+), 3 deletions(-). The budget is met exactly; the wording was cut three times to reach it (measured by wrapping each candidate at the file's 78-column prose width before editing).

Gates — each line is the gate's own verdict, re-run on the final commit 7bcfefc

check-skills-paths ✅ OK (94/95 stated path(s) resolve across 18 guide file(s); 1 baselined).
check-control-bytes ✅ OK (scanned 5714 tracked text file(s); skipped 85 binary).
check:doc-fences ✅ every TypeScript block in 223 document(s) is fenced ts/tsx/typescript ...
check-doc-links Links are valid across 17 scan roots.
check-changeset-presence ✅ No source of a released package changed in this range, so no changeset is owed.
check-doc-component-types ✅ Every documented component type is registered.
check-shell-escape-residue ✅ OK (4/4 root(s) resolved -- ... skills: 18 file(s), 235 fence(s) ...).
vitest (4 files) Test Files 4 passed (4) · Tests 107 passed (107)

The vitest set is every test file that reads this corpus: packages/components/src/__tests__/skill-guide-data-table-binding.test.tsx, scripts/__tests__/check-skills-paths.test.ts, scripts/__tests__/check-control-bytes.test.ts, scripts/__tests__/doc-version-claims.test.ts.

check-skills-paths got a positive control rather than a bare green: calling its own exported extractPathTokens on the edited file returns the new token — { line: 310, token: 'packages/core/src/evaluator/rowPredicateCanon.ts', pattern: false } — so the 94/95 verdict demonstrably includes the path this PR added.

No changeset: skills/** is not published source of a released package, and the gate says so in its own words above.

Two declared narrowings

check:doc-snippets was NOT RUN. Its own exit is 2 — "PRECONDITION NOT MET ... This is 'I could not run', NOT 'I ran and found errors'" — because it needs 21 package build closures, a repo-wide build this shared box cannot host in a foreground window. Instead its judged population was measured directly through its own exported extractor, scanFences:

• population read from the gate itself, not guessed: scanFences collects only ts/tsx fenced blocks; this file has exactly 1, at line 250 — above the edited hunk (307-313).
• count from that output: 1 block at base aca70f6 and 1 at head, same fenceLine 250, bodies byte-identical.
• counter-probe: appending a line to that block's body does change the extracted population, so "identical" is a measurement that could have failed rather than a vacuous pass.

pnpm lint was narrowed to nothing, on eslint's own answer.eslint --no-inline-config --format json on the changed file reports "File ignored because no matching configuration was supplied" — the repo's flat config matches no .md file at all, so the file count eslint judges in this diff is 0. No type-aware linting is configured, and the diff is one markdown file, so no untouched file's verdict can move either.

CI runs the full farm regardless; these two are the cheap local half, declared rather than silently skipped.

Governed-face handling

Left as draft for the PM's review chain — not flipped ready, not queued, no auto-merge. Note for the record: this repo's AGENTS.md §"受管面" explicitly places root skills/**outside the governed surface (only .claude/**, AGENTS.md, CLAUDE.md, docs/adr/** are governed there) and would permit self-merge on green. The dispatch treats the published skills corpus as governed for this fleet's review chain, which is the stricter of the two, so draft-only satisfies both. Flagged so the divergence is visible rather than assumed.

---

Generated by Claude Code

[zhuangjianguo]

Fixes#5759

Prose-only annotation of the three-way row-predicate sentence in the published skills guide skills/objectui/guides/schema-expressions.md (List-view/CEL tier). It presented record.status, bare status and data.status as peers, labelling record.* merely "canonical" — a word that reads as a style preference — while since Phase 1 (#5330 / PR #5737) the other two emit a one-time dev-console deprecation warning and retire once a stored-metadata survey sizes the window.

What changed

One sentence, at :307. record.* is now stated as the canon; bare status and data.status are stated as deprecated here (still bound, warned once in dev, retiring after the survey), pointing at packages/core/src/evaluator/rowPredicateCanon.ts; and the card's sharper point is carried: data.* is the trap — the server's authoring oracle accepts it silently and it then binds nothing at runtime, a constant false rather than an error.

"Deprecated here" is deliberate, not hedging: rowPredicateCanon.ts scopes the deprecation to the runtime record layer, because data is the canonical root of a metadata-editing form (ADR-0089 D3). An unqualified sentence in this corpus would contradict that.

No example was rewritten — the nearby data-model-tier and conditional-formatting examples are already record.*.

State verified on this branch, not taken from the card

The card was filed 2026-08-23; the words had to describe what is measurable now. Read at base aca70f6:

• listConditional.ts:273 still binds all three — { ...(opts.scope ?? {}), ...rowObj, data: rowObj, record: rowObj }. No spelling has been retired or turned into a refusal, so "still bound" is accurate and "retiring after a stored-metadata survey" is the correct tense.
• listConditional.ts:317 calls warnNonCanonicalRowSpelling(predicateText, rowObj, !opts.rowless, opts.label) on the CEL path, so both non-canonical spellings do warn on this tier, and data names the row here (not rowless), which is the condition the detector needs.
• rowPredicateCanon.ts still reads Phase 1: "the binding is UNCHANGED and every spelling still resolves", warn-once, "removable only after a stored-metadata survey".

Phase 2 (#5741) has not advanced past "warned".

Three-engine boundary held

Only the List-view/CEL row-predicate tier changed. The ${…}/On-suffix tier earlier in the file (SafeExpressionParser, where data.* is correct) and the flow tier (isFieldVisibleWhen, previews/screen-spec.ts) are untouched. The one-time-warning sentence at :316-317 belongs to legacy-syntax routing, a different mechanism from these spellings, and is untouched — the diff hunk stops above it.

Line-count readings (2026-08-21 whole-package ruling)

PM-set budget for this card: +4 lines max net on the file.

reading	before	after	net
skills/objectui/guides/schema-expressions.md (whole file)	621	625	+4
published bundle — all 18 .md under skills/	5682	5686	+4

git diff --stat: 1 file changed, 7 insertions(+), 3 deletions(-). The budget is met exactly; the wording was cut three times to reach it (measured by wrapping each candidate at the file's 78-column prose width before editing).

Gates — each line is the gate's own verdict, re-run on the final commit 7bcfefc

check-skills-paths ✅ OK (94/95 stated path(s) resolve across 18 guide file(s); 1 baselined).
check-control-bytes ✅ OK (scanned 5714 tracked text file(s); skipped 85 binary).
check:doc-fences ✅ every TypeScript block in 223 document(s) is fenced ts/tsx/typescript ...
check-doc-links Links are valid across 17 scan roots.
check-changeset-presence ✅ No source of a released package changed in this range, so no changeset is owed.
check-doc-component-types ✅ Every documented component type is registered.
check-shell-escape-residue ✅ OK (4/4 root(s) resolved -- ... skills: 18 file(s), 235 fence(s) ...).
vitest (4 files) Test Files 4 passed (4) · Tests 107 passed (107)

The vitest set is every test file that reads this corpus: packages/components/src/__tests__/skill-guide-data-table-binding.test.tsx, scripts/__tests__/check-skills-paths.test.ts, scripts/__tests__/check-control-bytes.test.ts, scripts/__tests__/doc-version-claims.test.ts.

check-skills-paths got a positive control rather than a bare green: calling its own exported extractPathTokens on the edited file returns the new token — { line: 310, token: 'packages/core/src/evaluator/rowPredicateCanon.ts', pattern: false } — so the 94/95 verdict demonstrably includes the path this PR added.

No changeset: skills/** is not published source of a released package, and the gate says so in its own words above.

Two declared narrowings

check:doc-snippets was NOT RUN. Its own exit is 2 — "PRECONDITION NOT MET ... This is 'I could not run', NOT 'I ran and found errors'" — because it needs 21 package build closures, a repo-wide build this shared box cannot host in a foreground window. Instead its judged population was measured directly through its own exported extractor, scanFences:

• population read from the gate itself, not guessed: scanFences collects only ts/tsx fenced blocks; this file has exactly 1, at line 250 — above the edited hunk (307-313).
• count from that output: 1 block at base aca70f6 and 1 at head, same fenceLine 250, bodies byte-identical.
• counter-probe: appending a line to that block's body does change the extracted population, so "identical" is a measurement that could have failed rather than a vacuous pass.

pnpm lint was narrowed to nothing, on eslint's own answer.eslint --no-inline-config --format json on the changed file reports "File ignored because no matching configuration was supplied" — the repo's flat config matches no .md file at all, so the file count eslint judges in this diff is 0. No type-aware linting is configured, and the diff is one markdown file, so no untouched file's verdict can move either.

CI runs the full farm regardless; these two are the cheap local half, declared rather than silently skipped.

Governed-face handling

Left as draft for the PM's review chain — not flipped ready, not queued, no auto-merge. Note for the record: this repo's AGENTS.md §"受管面" explicitly places root skills/**outside the governed surface (only .claude/**, AGENTS.md, CLAUDE.md, docs/adr/** are governed there) and would permit self-merge on green. The dispatch treats the published skills corpus as governed for this fleet's review chain, which is the stricter of the two, so draft-only satisfies both. Flagged so the divergence is visible rather than assumed.

---

Generated by Claude Code

[zhuangjianguo]

Fixes#5759

Prose-only annotation of the three-way row-predicate sentence in the published skills guide skills/objectui/guides/schema-expressions.md (List-view/CEL tier). It presented record.status, bare status and data.status as peers, labelling record.* merely "canonical" — a word that reads as a style preference — while since Phase 1 (#5330 / PR #5737) the other two emit a one-time dev-console deprecation warning and retire once a stored-metadata survey sizes the window.

What changed

One sentence, at :307. record.* is now stated as the canon; bare status and data.status are stated as deprecated here (still bound, warned once in dev, retiring after the survey), pointing at packages/core/src/evaluator/rowPredicateCanon.ts; and the card's sharper point is carried: data.* is the trap — the server's authoring oracle accepts it silently and it then binds nothing at runtime, a constant false rather than an error.

"Deprecated here" is deliberate, not hedging: rowPredicateCanon.ts scopes the deprecation to the runtime record layer, because data is the canonical root of a metadata-editing form (ADR-0089 D3). An unqualified sentence in this corpus would contradict that.

No example was rewritten — the nearby data-model-tier and conditional-formatting examples are already record.*.

State verified on this branch, not taken from the card

The card was filed 2026-08-23; the words had to describe what is measurable now. Read at base aca70f6:

• listConditional.ts:273 still binds all three — { ...(opts.scope ?? {}), ...rowObj, data: rowObj, record: rowObj }. No spelling has been retired or turned into a refusal, so "still bound" is accurate and "retiring after a stored-metadata survey" is the correct tense.
• listConditional.ts:317 calls warnNonCanonicalRowSpelling(predicateText, rowObj, !opts.rowless, opts.label) on the CEL path, so both non-canonical spellings do warn on this tier, and data names the row here (not rowless), which is the condition the detector needs.
• rowPredicateCanon.ts still reads Phase 1: "the binding is UNCHANGED and every spelling still resolves", warn-once, "removable only after a stored-metadata survey".

Phase 2 (#5741) has not advanced past "warned".

Three-engine boundary held

Only the List-view/CEL row-predicate tier changed. The ${…}/On-suffix tier earlier in the file (SafeExpressionParser, where data.* is correct) and the flow tier (isFieldVisibleWhen, previews/screen-spec.ts) are untouched. The one-time-warning sentence at :316-317 belongs to legacy-syntax routing, a different mechanism from these spellings, and is untouched — the diff hunk stops above it.

Line-count readings (2026-08-21 whole-package ruling)

PM-set budget for this card: +4 lines max net on the file.

reading	before	after	net
skills/objectui/guides/schema-expressions.md (whole file)	621	625	+4
published bundle — all 18 .md under skills/	5682	5686	+4

git diff --stat: 1 file changed, 7 insertions(+), 3 deletions(-). The budget is met exactly; the wording was cut three times to reach it (measured by wrapping each candidate at the file's 78-column prose width before editing).

Gates — each line is the gate's own verdict, re-run on the final commit 7bcfefc

check-skills-paths ✅ OK (94/95 stated path(s) resolve across 18 guide file(s); 1 baselined).
check-control-bytes ✅ OK (scanned 5714 tracked text file(s); skipped 85 binary).
check:doc-fences ✅ every TypeScript block in 223 document(s) is fenced ts/tsx/typescript ...
check-doc-links Links are valid across 17 scan roots.
check-changeset-presence ✅ No source of a released package changed in this range, so no changeset is owed.
check-doc-component-types ✅ Every documented component type is registered.
check-shell-escape-residue ✅ OK (4/4 root(s) resolved -- ... skills: 18 file(s), 235 fence(s) ...).
vitest (4 files) Test Files 4 passed (4) · Tests 107 passed (107)

The vitest set is every test file that reads this corpus: packages/components/src/__tests__/skill-guide-data-table-binding.test.tsx, scripts/__tests__/check-skills-paths.test.ts, scripts/__tests__/check-control-bytes.test.ts, scripts/__tests__/doc-version-claims.test.ts.

check-skills-paths got a positive control rather than a bare green: calling its own exported extractPathTokens on the edited file returns the new token — { line: 310, token: 'packages/core/src/evaluator/rowPredicateCanon.ts', pattern: false } — so the 94/95 verdict demonstrably includes the path this PR added.

No changeset: skills/** is not published source of a released package, and the gate says so in its own words above.

Two declared narrowings

check:doc-snippets was NOT RUN. Its own exit is 2 — "PRECONDITION NOT MET ... This is 'I could not run', NOT 'I ran and found errors'" — because it needs 21 package build closures, a repo-wide build this shared box cannot host in a foreground window. Instead its judged population was measured directly through its own exported extractor, scanFences:

• population read from the gate itself, not guessed: scanFences collects only ts/tsx fenced blocks; this file has exactly 1, at line 250 — above the edited hunk (307-313).
• count from that output: 1 block at base aca70f6 and 1 at head, same fenceLine 250, bodies byte-identical.
• counter-probe: appending a line to that block's body does change the extracted population, so "identical" is a measurement that could have failed rather than a vacuous pass.

pnpm lint was narrowed to nothing, on eslint's own answer.eslint --no-inline-config --format json on the changed file reports "File ignored because no matching configuration was supplied" — the repo's flat config matches no .md file at all, so the file count eslint judges in this diff is 0. No type-aware linting is configured, and the diff is one markdown file, so no untouched file's verdict can move either.

CI runs the full farm regardless; these two are the cheap local half, declared rather than silently skipped.

Governed-face handling

Left as draft for the PM's review chain — not flipped ready, not queued, no auto-merge. Note for the record: this repo's AGENTS.md §"受管面" explicitly places root skills/**outside the governed surface (only .claude/**, AGENTS.md, CLAUDE.md, docs/adr/** are governed there) and would permit self-merge on green. The dispatch treats the published skills corpus as governed for this fleet's review chain, which is the stricter of the two, so draft-only satisfies both. Flagged so the divergence is visible rather than assumed.

---

Generated by Claude Code

[zhuangjianguo]

Fixes#5759

Prose-only annotation of the three-way row-predicate sentence in the published skills guide skills/objectui/guides/schema-expressions.md (List-view/CEL tier). It presented record.status, bare status and data.status as peers, labelling record.* merely "canonical" — a word that reads as a style preference — while since Phase 1 (#5330 / PR #5737) the other two emit a one-time dev-console deprecation warning and retire once a stored-metadata survey sizes the window.

What changed

One sentence, at :307. record.* is now stated as the canon; bare status and data.status are stated as deprecated here (still bound, warned once in dev, retiring after the survey), pointing at packages/core/src/evaluator/rowPredicateCanon.ts; and the card's sharper point is carried: data.* is the trap — the server's authoring oracle accepts it silently and it then binds nothing at runtime, a constant false rather than an error.

"Deprecated here" is deliberate, not hedging: rowPredicateCanon.ts scopes the deprecation to the runtime record layer, because data is the canonical root of a metadata-editing form (ADR-0089 D3). An unqualified sentence in this corpus would contradict that.

No example was rewritten — the nearby data-model-tier and conditional-formatting examples are already record.*.

State verified on this branch, not taken from the card

The card was filed 2026-08-23; the words had to describe what is measurable now. Read at base aca70f6:

• listConditional.ts:273 still binds all three — { ...(opts.scope ?? {}), ...rowObj, data: rowObj, record: rowObj }. No spelling has been retired or turned into a refusal, so "still bound" is accurate and "retiring after a stored-metadata survey" is the correct tense.
• listConditional.ts:317 calls warnNonCanonicalRowSpelling(predicateText, rowObj, !opts.rowless, opts.label) on the CEL path, so both non-canonical spellings do warn on this tier, and data names the row here (not rowless), which is the condition the detector needs.
• rowPredicateCanon.ts still reads Phase 1: "the binding is UNCHANGED and every spelling still resolves", warn-once, "removable only after a stored-metadata survey".

Phase 2 (#5741) has not advanced past "warned".

Three-engine boundary held

Only the List-view/CEL row-predicate tier changed. The ${…}/On-suffix tier earlier in the file (SafeExpressionParser, where data.* is correct) and the flow tier (isFieldVisibleWhen, previews/screen-spec.ts) are untouched. The one-time-warning sentence at :316-317 belongs to legacy-syntax routing, a different mechanism from these spellings, and is untouched — the diff hunk stops above it.

Line-count readings (2026-08-21 whole-package ruling)

PM-set budget for this card: +4 lines max net on the file.

reading	before	after	net
skills/objectui/guides/schema-expressions.md (whole file)	621	625	+4
published bundle — all 18 .md under skills/	5682	5686	+4

git diff --stat: 1 file changed, 7 insertions(+), 3 deletions(-). The budget is met exactly; the wording was cut three times to reach it (measured by wrapping each candidate at the file's 78-column prose width before editing).

Gates — each line is the gate's own verdict, re-run on the final commit 7bcfefc

check-skills-paths ✅ OK (94/95 stated path(s) resolve across 18 guide file(s); 1 baselined).
check-control-bytes ✅ OK (scanned 5714 tracked text file(s); skipped 85 binary).
check:doc-fences ✅ every TypeScript block in 223 document(s) is fenced ts/tsx/typescript ...
check-doc-links Links are valid across 17 scan roots.
check-changeset-presence ✅ No source of a released package changed in this range, so no changeset is owed.
check-doc-component-types ✅ Every documented component type is registered.
check-shell-escape-residue ✅ OK (4/4 root(s) resolved -- ... skills: 18 file(s), 235 fence(s) ...).
vitest (4 files) Test Files 4 passed (4) · Tests 107 passed (107)

The vitest set is every test file that reads this corpus: packages/components/src/__tests__/skill-guide-data-table-binding.test.tsx, scripts/__tests__/check-skills-paths.test.ts, scripts/__tests__/check-control-bytes.test.ts, scripts/__tests__/doc-version-claims.test.ts.

check-skills-paths got a positive control rather than a bare green: calling its own exported extractPathTokens on the edited file returns the new token — { line: 310, token: 'packages/core/src/evaluator/rowPredicateCanon.ts', pattern: false } — so the 94/95 verdict demonstrably includes the path this PR added.

No changeset: skills/** is not published source of a released package, and the gate says so in its own words above.

Two declared narrowings

check:doc-snippets was NOT RUN. Its own exit is 2 — "PRECONDITION NOT MET ... This is 'I could not run', NOT 'I ran and found errors'" — because it needs 21 package build closures, a repo-wide build this shared box cannot host in a foreground window. Instead its judged population was measured directly through its own exported extractor, scanFences:

• population read from the gate itself, not guessed: scanFences collects only ts/tsx fenced blocks; this file has exactly 1, at line 250 — above the edited hunk (307-313).
• count from that output: 1 block at base aca70f6 and 1 at head, same fenceLine 250, bodies byte-identical.
• counter-probe: appending a line to that block's body does change the extracted population, so "identical" is a measurement that could have failed rather than a vacuous pass.

pnpm lint was narrowed to nothing, on eslint's own answer.eslint --no-inline-config --format json on the changed file reports "File ignored because no matching configuration was supplied" — the repo's flat config matches no .md file at all, so the file count eslint judges in this diff is 0. No type-aware linting is configured, and the diff is one markdown file, so no untouched file's verdict can move either.

CI runs the full farm regardless; these two are the cheap local half, declared rather than silently skipped.

Governed-face handling

Left as draft for the PM's review chain — not flipped ready, not queued, no auto-merge. Note for the record: this repo's AGENTS.md §"受管面" explicitly places root skills/**outside the governed surface (only .claude/**, AGENTS.md, CLAUDE.md, docs/adr/** are governed there) and would permit self-merge on green. The dispatch treats the published skills corpus as governed for this fleet's review chain, which is the stricter of the two, so draft-only satisfies both. Flagged so the divergence is visible rather than assumed.

---

Generated by Claude Code

[zhuangjianguo]

Fixes#5759

Prose-only annotation of the three-way row-predicate sentence in the published skills guide skills/objectui/guides/schema-expressions.md (List-view/CEL tier). It presented record.status, bare status and data.status as peers, labelling record.* merely "canonical" — a word that reads as a style preference — while since Phase 1 (#5330 / PR #5737) the other two emit a one-time dev-console deprecation warning and retire once a stored-metadata survey sizes the window.

What changed

One sentence, at :307. record.* is now stated as the canon; bare status and data.status are stated as deprecated here (still bound, warned once in dev, retiring after the survey), pointing at packages/core/src/evaluator/rowPredicateCanon.ts; and the card's sharper point is carried: data.* is the trap — the server's authoring oracle accepts it silently and it then binds nothing at runtime, a constant false rather than an error.

"Deprecated here" is deliberate, not hedging: rowPredicateCanon.ts scopes the deprecation to the runtime record layer, because data is the canonical root of a metadata-editing form (ADR-0089 D3). An unqualified sentence in this corpus would contradict that.

No example was rewritten — the nearby data-model-tier and conditional-formatting examples are already record.*.

State verified on this branch, not taken from the card

The card was filed 2026-08-23; the words had to describe what is measurable now. Read at base aca70f6:

• listConditional.ts:273 still binds all three — { ...(opts.scope ?? {}), ...rowObj, data: rowObj, record: rowObj }. No spelling has been retired or turned into a refusal, so "still bound" is accurate and "retiring after a stored-metadata survey" is the correct tense.
• listConditional.ts:317 calls warnNonCanonicalRowSpelling(predicateText, rowObj, !opts.rowless, opts.label) on the CEL path, so both non-canonical spellings do warn on this tier, and data names the row here (not rowless), which is the condition the detector needs.
• rowPredicateCanon.ts still reads Phase 1: "the binding is UNCHANGED and every spelling still resolves", warn-once, "removable only after a stored-metadata survey".

Phase 2 (#5741) has not advanced past "warned".

Three-engine boundary held

Only the List-view/CEL row-predicate tier changed. The ${…}/On-suffix tier earlier in the file (SafeExpressionParser, where data.* is correct) and the flow tier (isFieldVisibleWhen, previews/screen-spec.ts) are untouched. The one-time-warning sentence at :316-317 belongs to legacy-syntax routing, a different mechanism from these spellings, and is untouched — the diff hunk stops above it.

Line-count readings (2026-08-21 whole-package ruling)

PM-set budget for this card: +4 lines max net on the file.

reading	before	after	net
skills/objectui/guides/schema-expressions.md (whole file)	621	625	+4
published bundle — all 18 .md under skills/	5682	5686	+4

git diff --stat: 1 file changed, 7 insertions(+), 3 deletions(-). The budget is met exactly; the wording was cut three times to reach it (measured by wrapping each candidate at the file's 78-column prose width before editing).

Gates — each line is the gate's own verdict, re-run on the final commit 7bcfefc

check-skills-paths ✅ OK (94/95 stated path(s) resolve across 18 guide file(s); 1 baselined).
check-control-bytes ✅ OK (scanned 5714 tracked text file(s); skipped 85 binary).
check:doc-fences ✅ every TypeScript block in 223 document(s) is fenced ts/tsx/typescript ...
check-doc-links Links are valid across 17 scan roots.
check-changeset-presence ✅ No source of a released package changed in this range, so no changeset is owed.
check-doc-component-types ✅ Every documented component type is registered.
check-shell-escape-residue ✅ OK (4/4 root(s) resolved -- ... skills: 18 file(s), 235 fence(s) ...).
vitest (4 files) Test Files 4 passed (4) · Tests 107 passed (107)

The vitest set is every test file that reads this corpus: packages/components/src/__tests__/skill-guide-data-table-binding.test.tsx, scripts/__tests__/check-skills-paths.test.ts, scripts/__tests__/check-control-bytes.test.ts, scripts/__tests__/doc-version-claims.test.ts.

check-skills-paths got a positive control rather than a bare green: calling its own exported extractPathTokens on the edited file returns the new token — { line: 310, token: 'packages/core/src/evaluator/rowPredicateCanon.ts', pattern: false } — so the 94/95 verdict demonstrably includes the path this PR added.

No changeset: skills/** is not published source of a released package, and the gate says so in its own words above.

Two declared narrowings

check:doc-snippets was NOT RUN. Its own exit is 2 — "PRECONDITION NOT MET ... This is 'I could not run', NOT 'I ran and found errors'" — because it needs 21 package build closures, a repo-wide build this shared box cannot host in a foreground window. Instead its judged population was measured directly through its own exported extractor, scanFences:

• population read from the gate itself, not guessed: scanFences collects only ts/tsx fenced blocks; this file has exactly 1, at line 250 — above the edited hunk (307-313).
• count from that output: 1 block at base aca70f6 and 1 at head, same fenceLine 250, bodies byte-identical.
• counter-probe: appending a line to that block's body does change the extracted population, so "identical" is a measurement that could have failed rather than a vacuous pass.

pnpm lint was narrowed to nothing, on eslint's own answer.eslint --no-inline-config --format json on the changed file reports "File ignored because no matching configuration was supplied" — the repo's flat config matches no .md file at all, so the file count eslint judges in this diff is 0. No type-aware linting is configured, and the diff is one markdown file, so no untouched file's verdict can move either.

CI runs the full farm regardless; these two are the cheap local half, declared rather than silently skipped.

Governed-face handling

Left as draft for the PM's review chain — not flipped ready, not queued, no auto-merge. Note for the record: this repo's AGENTS.md §"受管面" explicitly places root skills/**outside the governed surface (only .claude/**, AGENTS.md, CLAUDE.md, docs/adr/** are governed there) and would permit self-merge on green. The dispatch treats the published skills corpus as governed for this fleet's review chain, which is the stricter of the two, so draft-only satisfies both. Flagged so the divergence is visible rather than assumed.

---

Generated by Claude Code

[zhuangjianguo]

Fixes#5759

Prose-only annotation of the three-way row-predicate sentence in the published skills guide skills/objectui/guides/schema-expressions.md (List-view/CEL tier). It presented record.status, bare status and data.status as peers, labelling record.* merely "canonical" — a word that reads as a style preference — while since Phase 1 (#5330 / PR #5737) the other two emit a one-time dev-console deprecation warning and retire once a stored-metadata survey sizes the window.

What changed

One sentence, at :307. record.* is now stated as the canon; bare status and data.status are stated as deprecated here (still bound, warned once in dev, retiring after the survey), pointing at packages/core/src/evaluator/rowPredicateCanon.ts; and the card's sharper point is carried: data.* is the trap — the server's authoring oracle accepts it silently and it then binds nothing at runtime, a constant false rather than an error.

"Deprecated here" is deliberate, not hedging: rowPredicateCanon.ts scopes the deprecation to the runtime record layer, because data is the canonical root of a metadata-editing form (ADR-0089 D3). An unqualified sentence in this corpus would contradict that.

No example was rewritten — the nearby data-model-tier and conditional-formatting examples are already record.*.

State verified on this branch, not taken from the card

The card was filed 2026-08-23; the words had to describe what is measurable now. Read at base aca70f6:

• listConditional.ts:273 still binds all three — { ...(opts.scope ?? {}), ...rowObj, data: rowObj, record: rowObj }. No spelling has been retired or turned into a refusal, so "still bound" is accurate and "retiring after a stored-metadata survey" is the correct tense.
• listConditional.ts:317 calls warnNonCanonicalRowSpelling(predicateText, rowObj, !opts.rowless, opts.label) on the CEL path, so both non-canonical spellings do warn on this tier, and data names the row here (not rowless), which is the condition the detector needs.
• rowPredicateCanon.ts still reads Phase 1: "the binding is UNCHANGED and every spelling still resolves", warn-once, "removable only after a stored-metadata survey".

Phase 2 (#5741) has not advanced past "warned".

Three-engine boundary held

Only the List-view/CEL row-predicate tier changed. The ${…}/On-suffix tier earlier in the file (SafeExpressionParser, where data.* is correct) and the flow tier (isFieldVisibleWhen, previews/screen-spec.ts) are untouched. The one-time-warning sentence at :316-317 belongs to legacy-syntax routing, a different mechanism from these spellings, and is untouched — the diff hunk stops above it.

Line-count readings (2026-08-21 whole-package ruling)

PM-set budget for this card: +4 lines max net on the file.

reading	before	after	net
skills/objectui/guides/schema-expressions.md (whole file)	621	625	+4
published bundle — all 18 .md under skills/	5682	5686	+4

git diff --stat: 1 file changed, 7 insertions(+), 3 deletions(-). The budget is met exactly; the wording was cut three times to reach it (measured by wrapping each candidate at the file's 78-column prose width before editing).

Gates — each line is the gate's own verdict, re-run on the final commit 7bcfefc

check-skills-paths ✅ OK (94/95 stated path(s) resolve across 18 guide file(s); 1 baselined).
check-control-bytes ✅ OK (scanned 5714 tracked text file(s); skipped 85 binary).
check:doc-fences ✅ every TypeScript block in 223 document(s) is fenced ts/tsx/typescript ...
check-doc-links Links are valid across 17 scan roots.
check-changeset-presence ✅ No source of a released package changed in this range, so no changeset is owed.
check-doc-component-types ✅ Every documented component type is registered.
check-shell-escape-residue ✅ OK (4/4 root(s) resolved -- ... skills: 18 file(s), 235 fence(s) ...).
vitest (4 files) Test Files 4 passed (4) · Tests 107 passed (107)

The vitest set is every test file that reads this corpus: packages/components/src/__tests__/skill-guide-data-table-binding.test.tsx, scripts/__tests__/check-skills-paths.test.ts, scripts/__tests__/check-control-bytes.test.ts, scripts/__tests__/doc-version-claims.test.ts.

check-skills-paths got a positive control rather than a bare green: calling its own exported extractPathTokens on the edited file returns the new token — { line: 310, token: 'packages/core/src/evaluator/rowPredicateCanon.ts', pattern: false } — so the 94/95 verdict demonstrably includes the path this PR added.

No changeset: skills/** is not published source of a released package, and the gate says so in its own words above.

Two declared narrowings

check:doc-snippets was NOT RUN. Its own exit is 2 — "PRECONDITION NOT MET ... This is 'I could not run', NOT 'I ran and found errors'" — because it needs 21 package build closures, a repo-wide build this shared box cannot host in a foreground window. Instead its judged population was measured directly through its own exported extractor, scanFences:

• population read from the gate itself, not guessed: scanFences collects only ts/tsx fenced blocks; this file has exactly 1, at line 250 — above the edited hunk (307-313).
• count from that output: 1 block at base aca70f6 and 1 at head, same fenceLine 250, bodies byte-identical.
• counter-probe: appending a line to that block's body does change the extracted population, so "identical" is a measurement that could have failed rather than a vacuous pass.

pnpm lint was narrowed to nothing, on eslint's own answer.eslint --no-inline-config --format json on the changed file reports "File ignored because no matching configuration was supplied" — the repo's flat config matches no .md file at all, so the file count eslint judges in this diff is 0. No type-aware linting is configured, and the diff is one markdown file, so no untouched file's verdict can move either.

CI runs the full farm regardless; these two are the cheap local half, declared rather than silently skipped.

Governed-face handling

Left as draft for the PM's review chain — not flipped ready, not queued, no auto-merge. Note for the record: this repo's AGENTS.md §"受管面" explicitly places root skills/**outside the governed surface (only .claude/**, AGENTS.md, CLAUDE.md, docs/adr/** are governed there) and would permit self-merge on green. The dispatch treats the published skills corpus as governed for this fleet's review chain, which is the stricter of the two, so draft-only satisfies both. Flagged so the divergence is visible rather than assumed.

---

Generated by Claude Code

[zhuangjianguo]

Fixes#5759

Prose-only annotation of the three-way row-predicate sentence in the published skills guide skills/objectui/guides/schema-expressions.md (List-view/CEL tier). It presented record.status, bare status and data.status as peers, labelling record.* merely "canonical" — a word that reads as a style preference — while since Phase 1 (#5330 / PR #5737) the other two emit a one-time dev-console deprecation warning and retire once a stored-metadata survey sizes the window.

What changed

One sentence, at :307. record.* is now stated as the canon; bare status and data.status are stated as deprecated here (still bound, warned once in dev, retiring after the survey), pointing at packages/core/src/evaluator/rowPredicateCanon.ts; and the card's sharper point is carried: data.* is the trap — the server's authoring oracle accepts it silently and it then binds nothing at runtime, a constant false rather than an error.

"Deprecated here" is deliberate, not hedging: rowPredicateCanon.ts scopes the deprecation to the runtime record layer, because data is the canonical root of a metadata-editing form (ADR-0089 D3). An unqualified sentence in this corpus would contradict that.

No example was rewritten — the nearby data-model-tier and conditional-formatting examples are already record.*.

State verified on this branch, not taken from the card

The card was filed 2026-08-23; the words had to describe what is measurable now. Read at base aca70f6:

• listConditional.ts:273 still binds all three — { ...(opts.scope ?? {}), ...rowObj, data: rowObj, record: rowObj }. No spelling has been retired or turned into a refusal, so "still bound" is accurate and "retiring after a stored-metadata survey" is the correct tense.
• listConditional.ts:317 calls warnNonCanonicalRowSpelling(predicateText, rowObj, !opts.rowless, opts.label) on the CEL path, so both non-canonical spellings do warn on this tier, and data names the row here (not rowless), which is the condition the detector needs.
• rowPredicateCanon.ts still reads Phase 1: "the binding is UNCHANGED and every spelling still resolves", warn-once, "removable only after a stored-metadata survey".

Phase 2 (#5741) has not advanced past "warned".

Three-engine boundary held

Only the List-view/CEL row-predicate tier changed. The ${…}/On-suffix tier earlier in the file (SafeExpressionParser, where data.* is correct) and the flow tier (isFieldVisibleWhen, previews/screen-spec.ts) are untouched. The one-time-warning sentence at :316-317 belongs to legacy-syntax routing, a different mechanism from these spellings, and is untouched — the diff hunk stops above it.

Line-count readings (2026-08-21 whole-package ruling)

PM-set budget for this card: +4 lines max net on the file.

reading	before	after	net
skills/objectui/guides/schema-expressions.md (whole file)	621	625	+4
published bundle — all 18 .md under skills/	5682	5686	+4

git diff --stat: 1 file changed, 7 insertions(+), 3 deletions(-). The budget is met exactly; the wording was cut three times to reach it (measured by wrapping each candidate at the file's 78-column prose width before editing).

Gates — each line is the gate's own verdict, re-run on the final commit 7bcfefc

check-skills-paths ✅ OK (94/95 stated path(s) resolve across 18 guide file(s); 1 baselined).
check-control-bytes ✅ OK (scanned 5714 tracked text file(s); skipped 85 binary).
check:doc-fences ✅ every TypeScript block in 223 document(s) is fenced ts/tsx/typescript ...
check-doc-links Links are valid across 17 scan roots.
check-changeset-presence ✅ No source of a released package changed in this range, so no changeset is owed.
check-doc-component-types ✅ Every documented component type is registered.
check-shell-escape-residue ✅ OK (4/4 root(s) resolved -- ... skills: 18 file(s), 235 fence(s) ...).
vitest (4 files) Test Files 4 passed (4) · Tests 107 passed (107)

The vitest set is every test file that reads this corpus: packages/components/src/__tests__/skill-guide-data-table-binding.test.tsx, scripts/__tests__/check-skills-paths.test.ts, scripts/__tests__/check-control-bytes.test.ts, scripts/__tests__/doc-version-claims.test.ts.

check-skills-paths got a positive control rather than a bare green: calling its own exported extractPathTokens on the edited file returns the new token — { line: 310, token: 'packages/core/src/evaluator/rowPredicateCanon.ts', pattern: false } — so the 94/95 verdict demonstrably includes the path this PR added.

No changeset: skills/** is not published source of a released package, and the gate says so in its own words above.

Two declared narrowings

check:doc-snippets was NOT RUN. Its own exit is 2 — "PRECONDITION NOT MET ... This is 'I could not run', NOT 'I ran and found errors'" — because it needs 21 package build closures, a repo-wide build this shared box cannot host in a foreground window. Instead its judged population was measured directly through its own exported extractor, scanFences:

• population read from the gate itself, not guessed: scanFences collects only ts/tsx fenced blocks; this file has exactly 1, at line 250 — above the edited hunk (307-313).
• count from that output: 1 block at base aca70f6 and 1 at head, same fenceLine 250, bodies byte-identical.
• counter-probe: appending a line to that block's body does change the extracted population, so "identical" is a measurement that could have failed rather than a vacuous pass.

pnpm lint was narrowed to nothing, on eslint's own answer.eslint --no-inline-config --format json on the changed file reports "File ignored because no matching configuration was supplied" — the repo's flat config matches no .md file at all, so the file count eslint judges in this diff is 0. No type-aware linting is configured, and the diff is one markdown file, so no untouched file's verdict can move either.

CI runs the full farm regardless; these two are the cheap local half, declared rather than silently skipped.

Governed-face handling

Left as draft for the PM's review chain — not flipped ready, not queued, no auto-merge. Note for the record: this repo's AGENTS.md §"受管面" explicitly places root skills/**outside the governed surface (only .claude/**, AGENTS.md, CLAUDE.md, docs/adr/** are governed there) and would permit self-merge on green. The dispatch treats the published skills corpus as governed for this fleet's review chain, which is the stricter of the two, so draft-only satisfies both. Flagged so the divergence is visible rather than assumed.

---

Generated by Claude Code

[zhuangjianguo]

Fixes#5759

Prose-only annotation of the three-way row-predicate sentence in the published skills guide skills/objectui/guides/schema-expressions.md (List-view/CEL tier). It presented record.status, bare status and data.status as peers, labelling record.* merely "canonical" — a word that reads as a style preference — while since Phase 1 (#5330 / PR #5737) the other two emit a one-time dev-console deprecation warning and retire once a stored-metadata survey sizes the window.

What changed

One sentence, at :307. record.* is now stated as the canon; bare status and data.status are stated as deprecated here (still bound, warned once in dev, retiring after the survey), pointing at packages/core/src/evaluator/rowPredicateCanon.ts; and the card's sharper point is carried: data.* is the trap — the server's authoring oracle accepts it silently and it then binds nothing at runtime, a constant false rather than an error.

"Deprecated here" is deliberate, not hedging: rowPredicateCanon.ts scopes the deprecation to the runtime record layer, because data is the canonical root of a metadata-editing form (ADR-0089 D3). An unqualified sentence in this corpus would contradict that.

No example was rewritten — the nearby data-model-tier and conditional-formatting examples are already record.*.

State verified on this branch, not taken from the card

The card was filed 2026-08-23; the words had to describe what is measurable now. Read at base aca70f6:

• listConditional.ts:273 still binds all three — { ...(opts.scope ?? {}), ...rowObj, data: rowObj, record: rowObj }. No spelling has been retired or turned into a refusal, so "still bound" is accurate and "retiring after a stored-metadata survey" is the correct tense.
• listConditional.ts:317 calls warnNonCanonicalRowSpelling(predicateText, rowObj, !opts.rowless, opts.label) on the CEL path, so both non-canonical spellings do warn on this tier, and data names the row here (not rowless), which is the condition the detector needs.
• rowPredicateCanon.ts still reads Phase 1: "the binding is UNCHANGED and every spelling still resolves", warn-once, "removable only after a stored-metadata survey".

Phase 2 (#5741) has not advanced past "warned".

Three-engine boundary held

Only the List-view/CEL row-predicate tier changed. The ${…}/On-suffix tier earlier in the file (SafeExpressionParser, where data.* is correct) and the flow tier (isFieldVisibleWhen, previews/screen-spec.ts) are untouched. The one-time-warning sentence at :316-317 belongs to legacy-syntax routing, a different mechanism from these spellings, and is untouched — the diff hunk stops above it.

Line-count readings (2026-08-21 whole-package ruling)

PM-set budget for this card: +4 lines max net on the file.

reading	before	after	net
skills/objectui/guides/schema-expressions.md (whole file)	621	625	+4
published bundle — all 18 .md under skills/	5682	5686	+4

git diff --stat: 1 file changed, 7 insertions(+), 3 deletions(-). The budget is met exactly; the wording was cut three times to reach it (measured by wrapping each candidate at the file's 78-column prose width before editing).

Gates — each line is the gate's own verdict, re-run on the final commit 7bcfefc

check-skills-paths ✅ OK (94/95 stated path(s) resolve across 18 guide file(s); 1 baselined).
check-control-bytes ✅ OK (scanned 5714 tracked text file(s); skipped 85 binary).
check:doc-fences ✅ every TypeScript block in 223 document(s) is fenced ts/tsx/typescript ...
check-doc-links Links are valid across 17 scan roots.
check-changeset-presence ✅ No source of a released package changed in this range, so no changeset is owed.
check-doc-component-types ✅ Every documented component type is registered.
check-shell-escape-residue ✅ OK (4/4 root(s) resolved -- ... skills: 18 file(s), 235 fence(s) ...).
vitest (4 files) Test Files 4 passed (4) · Tests 107 passed (107)

The vitest set is every test file that reads this corpus: packages/components/src/__tests__/skill-guide-data-table-binding.test.tsx, scripts/__tests__/check-skills-paths.test.ts, scripts/__tests__/check-control-bytes.test.ts, scripts/__tests__/doc-version-claims.test.ts.

check-skills-paths got a positive control rather than a bare green: calling its own exported extractPathTokens on the edited file returns the new token — { line: 310, token: 'packages/core/src/evaluator/rowPredicateCanon.ts', pattern: false } — so the 94/95 verdict demonstrably includes the path this PR added.

No changeset: skills/** is not published source of a released package, and the gate says so in its own words above.

Two declared narrowings

check:doc-snippets was NOT RUN. Its own exit is 2 — "PRECONDITION NOT MET ... This is 'I could not run', NOT 'I ran and found errors'" — because it needs 21 package build closures, a repo-wide build this shared box cannot host in a foreground window. Instead its judged population was measured directly through its own exported extractor, scanFences:

• population read from the gate itself, not guessed: scanFences collects only ts/tsx fenced blocks; this file has exactly 1, at line 250 — above the edited hunk (307-313).
• count from that output: 1 block at base aca70f6 and 1 at head, same fenceLine 250, bodies byte-identical.
• counter-probe: appending a line to that block's body does change the extracted population, so "identical" is a measurement that could have failed rather than a vacuous pass.

pnpm lint was narrowed to nothing, on eslint's own answer.eslint --no-inline-config --format json on the changed file reports "File ignored because no matching configuration was supplied" — the repo's flat config matches no .md file at all, so the file count eslint judges in this diff is 0. No type-aware linting is configured, and the diff is one markdown file, so no untouched file's verdict can move either.

CI runs the full farm regardless; these two are the cheap local half, declared rather than silently skipped.

Governed-face handling

Left as draft for the PM's review chain — not flipped ready, not queued, no auto-merge. Note for the record: this repo's AGENTS.md §"受管面" explicitly places root skills/**outside the governed surface (only .claude/**, AGENTS.md, CLAUDE.md, docs/adr/** are governed there) and would permit self-merge on green. The dispatch treats the published skills corpus as governed for this fleet's review chain, which is the stricter of the two, so draft-only satisfies both. Flagged so the divergence is visible rather than assumed.

---

Generated by Claude Code