[spec] Field.inlineColumns is z.array(z.any()) — a mis-keyed inline-grid column publishes clean and renders blank cells

[yinlianghui]

Found while implementing objectui#3951 (grid columns converge on the declared name key). Out of that card's scope — filed, not fixed.

What

packages/spec/src/data/field.zod.ts:553 declares the inline master-detail grid's explicit column list with no shape at all:

inlineColumns: z.array(z.any()).optional().describe('Explicit columns for the inline grid (derived from the child object when omitted)'),

z.any() means every column object validates: right keys, wrong keys, misspelled keys, empty objects. Publish-time validation has nothing to check against, so the only place a mis-keyed column is ever noticed is the browser, as blank cells.

Why it matters now

objectui#3951 was exactly this failure with the divergence in the renderer instead of the metadata: GridColumnDefinition declared the column key as name while the widget read field, so metadata authored from the published type rendered a grid with the right row COUNT and every cell empty, plus a React missing-key warning. That renderer is now aligned to name (objectui PR for #3951).

What that fix cannot reach is the authoring surface. After it lands, a relationship field authored as

inlineEdit: 'grid',
inlineColumns: [{ field: 'amount', label: 'Amount', type: 'currency' }],

still parses green, still deploys, and still renders a grid of blank cells — because z.any() accepts the retired spelling and the renderer no longer reads it. The same holds for any other typo (lable, colmns, a bare string where an object belongs).

Census taken during objectui#3951: zero inlineColumns usages across examples/app-crm, examples/app-showcase, examples/app-todo — every inline grid in the example apps takes the auto-derived columns path. So nothing in-repo is broken today; the exposure is for downstream apps and for AI-authored metadata, which is the repo's stated second axis: a lenient producer schema is where generated metadata errors hide.

Suggested direction (for triage, not a decision)

Give inlineColumns a real element schema mirroring ObjectUI's GridColumnDefinition — name (required), label?, type?, width?, required?, plus the column extras the grid reads (options, reference, readonlyWhen, requiredWhen, expression, scale) — and make it strict, so an unknown key is a named rejection at publish time rather than a blank cell at render time. relatedListColumns (field.zod.ts:588) is z.array(z.any()) for the same reason and is worth checking in the same pass; note lookupColumns (:611) already has a typed element schema, so the pattern exists in this file.

Sizing the strictness is a spec decision (which extras are authorable, whether the renderer's column extras belong in the spec at all), which is why this is filed rather than fixed.

Reproduce

Author any object with a master_detail child whose relationship field sets inlineEdit: 'grid' and inlineColumns: [{ field: 'x' }], then FieldSchema.safeParse it — it succeeds. Swap field for a nonsense key such as zzz and it still succeeds.

[os-zhuang]

Triage: decision card — needs-user-decision + domain:spec (semantic face: a typed element schema narrows the acceptance set, i.e. accept/reject behavior changes; clause-② contract-review tier applies whenever this is dispatched), type Task. Rationale: the tightening direction is standing doctrine (strict producer schema over consumer tolerance), but the element shape is a real contract decision — which column extras (options, reference, readonlyWhen, requiredWhen, expression, scale) become authorable spec surface, and whether relatedListColumns (the same z.array(z.any()) shape at field.zod.ts:588) rides in the same pass. lookupColumns (:611) already has a typed element schema, so the in-file pattern exists. Dedup: inlineColumns matches only this card.

Four-prong analysis:

1. Platform long-term coherence — three sibling column lists in one file under two disciplines (typed vs z.any()); converging on the lookupColumns pattern shrinks the special case rather than growing one.
2. Measured business pull — zero in-repo (0 inlineColumns usages across the three example apps); the exposure is downstream apps and AI-authored metadata — and objectui#3951 just measured this exact failure (blank cells with the right row count) one seam over.
3. AI-agent error-resistance — textbook case: z.any() on an authoring surface is where generated-metadata errors hide; a strict element schema turns a mis-keyed column into a named publish-time rejection instead of blank cells at render time.
4. Startup scope discipline — smallest honest scope is name required plus the keys the renderer actually reads today, strict; declining to admit renderer extras into the spec is a legitimate outcome and should be decided, not defaulted.

Recommendation: approve tightening with a minimal strict element schema mirroring ObjectUI's GridColumnDefinition (name required), rule relatedListColumns in the same pass, and rule explicitly on whether the renderer's column extras enter the spec surface. Blast-radius note: example apps are clean, so the narrowing breaks no in-repo metadata.

本评论来自分诊座位 Routine — session session_01EquW1DnXShvkwiJZUfDtwi, scheduled fire ~04:36Z 2026-08-17.

---

Generated by Claude Code

[os-zhuang]

Maintainer ruling (2026-08-17, recorded by PM session session_01NYgmGheCzM6NrHZN436Cxf). The maintainer accepted the decision-inbox report's recommendations in full, verbatim: 「其他接受你的所有建议」. For this card:

Ruled: approved — inlineColumns gets a real, strict element schema.

• Shape mirrors ObjectUI's GridColumnDefinition: name (required), label?, type?, width?, required?, plus the column extras the grid actually reads (options, reference, readonlyWhen, requiredWhen, expression, scale) — the exact authorable-extras set is settled by the implementing dev against the renderer's measured reads, reported in the PR.
• Strict: an unknown key is a named rejection at publish time (the 未知键静默剥离仍是全仓默认:把 #3405 的 strict 收紧从一个 schema 推广到整个可授权面(ADR-0078 完整性闸门) #4001 refusal shape, prescription naming name when field is used), never a blank cell at render time.
• relatedListColumns (field.zod.ts:588) checked in the same pass — same z.array(z.any()) disease; lookupColumns (:611) is the in-file precedent to follow.
• Migration cost measured at zero in-repo (no inlineColumns usage in any example app); accept-set narrowing on a published authoring surface ⇒ claude-fable-5 mandatory, spec lane.

Label flipped needs-user-decision → pm:queue in the same stroke.

---

Generated by Claude Code

[os-steve]

Claim: PM loop round 5
Session: session_01Fs18A2DdXLVN2h8PaaFBcP
Branch: claude/issue-9227-inline-columns-strict
Worktree: objectstack-issue-9227
Domain: domain:spec
File surface: packages/spec/src/data/field.zod.ts (regions: inlineColumns at current :835 and relatedListColumns at :870 — card-body line numbers have drifted), field tests, generated baselines/forms the strictness change moves, .changeset/; objectui is read-only measurement input for the renderer's actual column reads (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: fable — mandatory per the maintainer's 2026-08-17 ruling on this card (accept-set narrowing on a published authoring surface); also the ⚠️ #8656 pre-dispatch check (Field builder-object edits ⇒ fable rider) is subsumed by the mandate
Serial constraints cleared: no open PR touches field.zod.ts (checked this round); #9176 / #9310 claimed this same round (this seat) touch metadata-protocol/protocol.ts+protocol.zod.ts and liveness/doc.json respectively — file-disjoint; generated-artifact family overlap possible at landing ⇒ serial relay applies there, not at development. No in-flight sibling's pins assert inlineColumns/relatedListColumns behaviour.

---

Generated by Claude Code

[os-steve]

Claim amendment (same session session_01Fs18A2DdXLVN2h8PaaFBcP, branch claude/issue-9227-inline-columns-strict): file surface extended with examples/app-showcase/src/data/objects/invoice.object.ts.

Census re-verification on current main (the ruling's precondition) found that authored inlineColumns usage HAS appeared since the issue's census: invoice.object.ts lines ~213-221 author seven column entries in the retired field spelling ({ field: 'product' } ...), landed 2026-08-16 alongside #9123. The narrowing this card orders refuses exactly that spelling, so the fixture is in this PR's fixture-triage radius (respell to the ruling's prescribed name key). Sequencing consequence vs the objectui pin will be reported in the dev report as an open question. No open PR touches this file (checked 2026-08-17).

---

Generated by Claude Code

---

Generated by Claude Code