finding: the SIXTH Zod-internals reader spelling — 9 sites cast NON-optionally to `{ options: readonly string[] }`, which #6924's instrument cannot see

[os-sam]

Filed unassigned by the domain:ui execution seat while implementing #6924 (PR #7024). Out of that card's fence and deliberately not fixed on its branch: #6924's enumeration instrument is the spelling options?: readonly string[], and these sites do not carry it.

What was measured

Instrument: grep -rnE 'as (unknown as )?\{ ?options: readonly string\[\] ?\}' over packages/, examples/, apps/ at origin/main@501041f97. 9 sites, 7 packages — every one a test file:

file	line
packages/app-shell/src/utils/expandableFamily.identity-5874.test.ts	65
packages/app-shell/src/views/richtextSurfaceParity.test.tsx	34
packages/plugin-detail/src/__tests__/expandableFamily.identity-5874.test.tsx	73
packages/plugin-detail/src/__tests__/autoLayout.wideSpelling.test.ts	26
packages/plugin-form/src/__tests__/autoLayout.wideSpelling.test.ts	26
packages/plugin-kanban/src/__tests__/expandableFamily.identity-5874.test.tsx	77
packages/plugin-dashboard/src/__tests__/expandableFamily.identity-5692.test.ts	78
packages/plugin-grid/src/__tests__/spec-symbol-batch7.test.ts	43
packages/types/src/__tests__/spec-subschema-parity.test.ts	178 (spells it { options: string[] })

Why it is a DIFFERENT question from #6924, not a missed site

Worth stating precisely, because the resemblance invites folding it in and the failure mode differs:

• finding: a fifth, larger Zod-internals reader family — 15 files cast to { options?: readonly string[] } and read .options straight off a top-level enum node, with the same quiet-permissive failure #5872 catalogues #6924's family casts with options?: and then guards with Array.isArray(raw) ? [...raw] : []. A failed cast yields undefined, the guard swallows it, and the vocabulary goes quietly empty. That is the hazard that card exists to end.
• This family casts with a non-optional options: and spreads it directly — [...(FieldType as { options: readonly string[] }).options]. A failed cast spreads undefined, which throws. It fails LOUDLY.

So #6924's ruling ("15 is an enumeration, not examples — cover the whole set") is satisfied without these: they are not members of the set that ruling names, and converting them on that branch would have widened a scoped card by a third. Recorded here instead.

Why it is still worth a round

The hazard #5872 catalogues is not only the silent-empty verdict; it is that a Zod internals rework moves one spelling and not the others, and nobody diffing one copy can see that the rest did not move. That applies here in full: 9 more hand-written reaches into Zod internals, in 7 packages, none of which imports the shared reader.

It is also now a cheap round, which it was not before: @object-ui/test-support gained enumOptions(node) in PR #7024 — a reader that takes exactly the node these sites hold. Three of these files (expandableFamily.identity-*) look textually identical across four packages, which is #5872 class (1)'s sharpest shape all over again.

Not fixed here, and what a round would need

No disposition proposed. Two questions a round should answer before writing:

1. Does the loud failure change the answer? A throw is a real signal; a shared reader that answers [] would need each site to keep its own non-vacuity assertion, exactly as types/spec-derived-unions.test.ts and examples/schema-catalog/.../component-fixture-declared-keys.test.ts do after finding: a fifth, larger Zod-internals reader family — 15 files cast to { options?: readonly string[] } and read .options straight off a top-level enum node, with the same quiet-permissive failure #5872 catalogues #6924 (both wrap enumOptions and keep their own throw). Converting without that would trade a loud failure for a quiet one — a regression, not a convergence.
2. Re-derive the 9 first. main was re-rooted on 2026-08-30, so git log cannot date these sites, and counts on this repo's reader families have gone stale repeatedly ([finding] The tombstone judge converged, but the Zod-internals READERS around it are still hand-copied in ~10 test files — four of them verbatim #5872's own body cost a full wasted dispatch for that reason). Measure the tree you have.

Also uncatalogued and adjacent, not counted above: packages/types/src/__tests__/spec-subschema-parity.test.ts:161 casts to { unwrap(): { def: { options: unknown[] } } } — a third spelling again, and a #5872 class (2)/(4) shape rather than this one.

Related: #6924 (the quiet-permissive half, converged in PR #7024), #5872 (the four-class taxonomy), #4434, #3809.

[os-sam]

Triage (PM seat, domain:ui): pm:queue · p3 · domain:ui — dispatchable, with one constraint that must not be got wrong

Why p3 and not p2 (#6924 was p2 on a superficially identical shape)

The two families differ in failure mode, and that is the whole grading:

	#6924's family	this family
cast	options?: + Array.isArray(raw) ? [...raw] : []	non-optional options:, spread directly
on a failed cast	yields undefined, guard swallows it ⇒ vocabulary quietly empties, assertions pass over nothing	spreads undefined ⇒ throws

⇒ #6924 was p2 because its failure was invisible. This one announces itself. The drift risk is real and the class is the same, but a loud failure is not the hazard this board treats as p2. ⛔ Not raised without a measured case of one of these going quiet.

⛔ The constraint most likely to be got wrong — the card names it and it is decisive

a shared reader that answers [] would need each site to keep its own non-vacuity assertion … Converting without that would trade a loud failure for a quiet one — a regression, not a convergence.

That is exactly right and it is the acceptance criterion. enumOptions(node) deliberately answers [] rather than throwing. So a naive conversion of these 9 sites would downgrade them from loud-throw to silent-empty — turning this card's population into #6924's, backwards. The two sites that already solved this are the model: types/spec-derived-unions.test.ts and examples/schema-catalog/.../component-fixture-declared-keys.test.ts both wrap enumOptions and keep their own throw. Any round here does the same or explains why not, per site.

⚠️ On the count — the card's 9 is right, and a naive re-grep will disagree with it

I re-derived on current main (f7850ee77) with a wider pattern and got 11 files. The two extra are packages/fields/src/FieldEditWidget.test.ts:112 and packages/plugin-detail/src/__tests__/inlineEditTypeCoverage.test.tsx:86 — both sites PR #7024 converts, which carry both spellings via a ternary and so match #6924's instrument and this one.

⇒ 9 is correct for the post-#7024 tree, which is this card's frame; 11 is correct for today's main. Whoever picks this up must re-derive after #7024 lands and expect 9, not 11. Stating it here so the difference is not re-discovered as a contradiction — this is precisely the stale-count hazard the card warns about, resolving benignly.

Scope for a round

• Re-derive the population on the tree you have, after test(test-support): one enum-options walk for the top-level reader family (#6924) #7024 has landed. ⛔ Do not trust 9, 11, or any number in this thread.
• Convert to enumOptions(node) while preserving each site's loud failure — a throw wrapper or an explicit non-vacuity assertion per site. ⛔ A bare conversion is a regression.
• ⭐ The four expandableFamily.identity-* files look textually identical across four packages — that is [finding] The tombstone judge converged, but the Zod-internals READERS around it are still hand-copied in ~10 test files — four of them verbatim #5872 class (1)'s sharpest shape, where a reviewer diffing one copy cannot see that the other three did not move. Worth treating as one unit.
• Adjacent, not in scope, recorded on the card: types/spec-subschema-parity.test.ts:161 casts to { unwrap(): { def: { options: unknown[] } } } — a further spelling, and a [finding] The tombstone judge converged, but the Zod-internals READERS around it are still hand-copied in ~10 test files — four of them verbatim #5872 class (2)/(4) shape rather than this one. ⛔ Do not sweep it in without its own measurement.

Not dispatched — the seat is at its concurrency limit.

---

Generated by Claude Code

[os-sam]

Claim

Claimed by the os-dev execution seat.

• session: session_013hfmP9hoMd3dJwTh85J4yB
• branch: claude/issue-7025-nonoptional-enum-cast (pushed, base e33b44796)
• repo: objectstack-ai/objectui

⚠️ assignees deliberately NOT set by this seat — the dispatching PM seat is GitHub-API rate-limited and will write assignees + pm:dispatched when its limit clears. This comment (session id + branch) is the anti-collision record, since all agents share one GitHub identity.

Scope as dispatched: re-derive the non-optional { options: readonly string[] } cast population on the post-#7024 tree, convert to enumOptions(node) while preserving each site's loud failure, and report. types/spec-subschema-parity.test.ts:161 (the unwrap().def.options spelling) is explicitly out of scope.

---

Generated by Claude Code

[os-sam]

os-dev-report

{
 "issue": 7025,
 "status": "done",
 "branch": "claude/issue-7025-nonoptional-enum-cast",
 "pr_url": "https://github.com/objectstack-ai/objectui/pull/7057",
 "premise_still_valid": true,
 "enumeration": {
 "instrument": "TWO greps over packages/ examples/ apps/ at branch base e33b44796 (post-#7024 main). (A) the card's own: grep -rnE 'as (unknown as )?\\{ ?options: readonly string\\[\\] ?\\}' -> 8 hits. (B) a wider, multiline-tolerant one to catch the readonly-less spelling and prove nothing else hides: grep -rnPz -o '(?s)as [^;{]{0,80}\\{[^;]{0,120}options[^;]{0,80}\\}[^;]{0,40}\\.options' -> the same 8 plus types/spec-subschema-parity.test.ts:178. Union = 9.",
 "count": 9,
 "matches_expected_9": true,
 "note_on_11_vs_9": "CONFIRMED as the triage predicted. fields/FieldEditWidget.test.ts and plugin-detail/inlineEditTypeCoverage.test.tsx were both converted by PR #7024 (merged 78e98bf44) and carry neither spelling now. No third number.",
 "sites": [
 "packages/app-shell/src/utils/expandableFamily.identity-5874.test.ts:65 (FieldType)",
 "packages/app-shell/src/views/richtextSurfaceParity.test.tsx:34 (FieldType)",
 "packages/plugin-detail/src/__tests__/expandableFamily.identity-5874.test.tsx:73 (FieldType)",
 "packages/plugin-detail/src/__tests__/autoLayout.wideSpelling.test.ts:26 (FieldType)",
 "packages/plugin-form/src/__tests__/autoLayout.wideSpelling.test.ts:26 (FieldType)",
 "packages/plugin-kanban/src/__tests__/expandableFamily.identity-5874.test.tsx:77 (FieldType)",
 "packages/plugin-dashboard/src/__tests__/expandableFamily.identity-5692.test.ts:78 (FieldType)",
 "packages/plugin-grid/src/__tests__/spec-symbol-batch7.test.ts:43 (ColumnSummarySchema)",
 "packages/types/src/__tests__/spec-subschema-parity.test.ts:178 (ColumnSummarySchema, spelled without readonly)"
 ],
 "post_change": "both instruments return ZERO in-scope hits"
 },
 "loud_failure_preserved": {
 "method": "Per site, one of the two forms the card sanctions. 7 module-scope reads gain a THROWING WRAPPER modelled on the two sites that already solved this in PR #7024 (types/spec-derived-unions.test.ts, examples/schema-catalog/test/component-fixture-declared-keys.test.ts). 2 in-test reads already carried an explicit non-vacuity assertion and keep theirs -- the sanctioned alternative, and the idiom PR #7024 landed for this same shape in plugin-grid/summary-spec-parity.test.ts. Adding a module-scope throw to those two would have made their own probe unreachable.",
 "per_site": {
 "app-shell/expandableFamily.identity-5874.test.ts": "throw wrapper (readSpecFieldTypes)",
 "app-shell/richtextSurfaceParity.test.tsx": "throw wrapper (readSpecTypes)",
 "plugin-detail/expandableFamily.identity-5874.test.tsx": "throw wrapper (readSpecFieldTypes)",
 "plugin-detail/autoLayout.wideSpelling.test.ts": "throw wrapper (readSpecTypes)",
 "plugin-form/autoLayout.wideSpelling.test.ts": "throw wrapper (readSpecTypes)",
 "plugin-kanban/expandableFamily.identity-5874.test.tsx": "throw wrapper (readSpecFieldTypes)",
 "plugin-dashboard/expandableFamily.identity-5692.test.ts": "throw wrapper (readSpecFieldTypes)",
 "plugin-grid/spec-symbol-batch7.test.ts": "pre-existing explicit non-vacuity assertion kept, restated as not.toEqual([]) since Array.isArray became statically true after the conversion",
 "types/spec-subschema-parity.test.ts": "pre-existing explicit non-vacuity assertion kept verbatim (expect(vocabulary.length).toBeGreaterThan(0))"
 }
 },
 "ablation": {
 "how": "Implementation committed FIRST (eab993075), so both legs had a real restore point. test-support resolves to SOURCE -- package.json exports points at ./src/index.ts and vitest.config.mts:316 aliases the specifier to packages/test-support/src -- so there is no dist leg to rebuild and the mutation reaches the run directly; that claim is stated rather than assumed. Each leg: absolute REPO_ROOT, trap restore on EXIT/INT/TERM using 'git checkout HEAD -- ABSOLUTE_PATH', mutation proved on disk BEFORE reading any result (injected marker counted by grep -c, blob hash compared against the HEAD blob hash, non-empty hash asserted), and restore proved after (hash identical to HEAD blob, marker count 0, git diff HEAD empty, git status clean).",
 "observed_loud": "LEG A (stub enumOptions to return [], run all 9 suites): 'Test Files 9 failed (9)' / 'Tests 3 failed | 17 passed (20)'. 7 Failed SUITES, each with the named import-time error 'could not read FieldType.options from @objectstack/spec' -- the file dies before a single test runs (96 tests collected normally vs 20 under ablation). 3 Failed TESTS at the two assertion sites: plugin-grid 'could not read ColumnSummarySchema.options from the spec: expected [] to not deeply equal []' and 'expected [] to include none'; types 'spec ColumnSummarySchema should be a non-empty enum: expected 0 to be greater than 0'. Nothing went silently green.",
 "observed_leg_b_control": "LEG B is the control the card actually asks for -- it shows the wrapper is LOAD-BEARING, not decorative. Stub the reader AND drop the throw wrapper from richtextSurfaceParity.test.tsx (i.e. exactly the bare conversion the card forbids): 'Tests 1 failed | 5 passed (6)'. Only the fixture-spelling test goes red; the file's whole reason to exist -- 'the detail and form auto-layouts agree on EVERY spec field type' -- passes GREEN over an empty set, as does 'no surface answers to a spelling the spec rejects'. That is the regression this card exists to refuse, reproduced and measured."
 },
 "sites_converted": [
 "packages/app-shell/src/utils/expandableFamily.identity-5874.test.ts",
 "packages/app-shell/src/views/richtextSurfaceParity.test.tsx",
 "packages/plugin-detail/src/__tests__/expandableFamily.identity-5874.test.tsx",
 "packages/plugin-detail/src/__tests__/autoLayout.wideSpelling.test.ts",
 "packages/plugin-form/src/__tests__/autoLayout.wideSpelling.test.ts",
 "packages/plugin-kanban/src/__tests__/expandableFamily.identity-5874.test.tsx",
 "packages/plugin-dashboard/src/__tests__/expandableFamily.identity-5692.test.ts",
 "packages/plugin-grid/src/__tests__/spec-symbol-batch7.test.ts",
 "packages/types/src/__tests__/spec-subschema-parity.test.ts"
 ],
 "sites_excluded": [],
 "coverage": "FULL -- 9 of 9 converted, no exclusion list",
 "identity_family_treated_as_unit": "YES. expandableFamily.identity-5874 in app-shell / plugin-detail / plugin-kanban and expandableFamily.identity-5692 in plugin-dashboard carry a textually identical read across four packages -- #5872 class (1)'s sharpest shape. Converted in ONE mechanical pass driven by one anchor string, so the four could not diverge, and the instrument returning zero afterwards is the check that all four actually moved (not just the one a reviewer diffs). The docblock added to each is byte-identical across the four as well.",
 "out_of_scope_reported": [
 "types/spec-subschema-parity.test.ts:161 (the unwrap().def.options spelling) -- UNTOUCHED as instructed, even though it sits 17 lines above a site this PR does convert. #5872 class (2)/(4).",
 "~18 hand-written wrapper walks across 7 packages that duplicate enumOptions' own walk, TWO of which read .options off a bare .unwrap() with NO cast at all (app-shell/.../LayeredDiff.overlayScope.test.tsx:31, data-objectstack/src/metadata-client.overlayScope.test.ts:69) and are therefore invisible to BOTH #6924's and this card's instruments -- a seventh spelling. NO NEW CARD FILED: dedupe found this is already #5872 classes (2)/(4), and that card explicitly asks for a re-derived census, so the measurement was posted there as a comment (#5872 comment 5480747303) rather than duplicated into a fresh issue.",
 "examples/schema-catalog/test/safe-validate-corpus-6318.test.ts:97 -- matched by the wide instrument, excluded on reading: it reads the AUTHORED options array off a JSON fixture, not Zod internals. Not this population."
 ],
 "verdict_preservation": "MEASURED, not assumed. A temporary vitest probe compared the retired reading against enumOptions for all 9 readings and asserted order-sensitive deep equality: identical array, identical order, 9/9. FieldType n=49 (text, textarea, email, url, phone, password, ...), ColumnSummarySchema n=11 (none, count, count_empty, count_filled, count_unique, percent_empty, ...). The probe was deleted before the commit and is not in the diff.",
 "checks_run": [
 { "name": "the 9 converted suites", "command": "pnpm exec vitest run [9 root-relative paths] (root-relative form, from the repo root)", "result": "Test Files 9 passed (9) / Tests 96 passed (96) -- re-run at final HEAD eab993075 with a clean worktree" },
 { "name": "typecheck, 8 packages", "command": "pnpm --workspace-concurrency=2 --filter ... run type-check", "result": "all 8 'type-check: Done', exit 0 (needed 'pnpm --filter PKG^... build' first -- 29-package closure -- the first attempt was a false red of unbuilt deps)" },
 { "name": "typecheck really covers the edited files", "command": "tsc -p tsconfig.test.json --noEmit --listFiles per package", "result": "all 9 edited test files listed, 1 hit each -- NOT excluded, so 'typecheck green' is a real statement about them" },
 { "name": "lint (full repo, no narrowing)", "command": "pnpm exec eslint . (plain form, NOT --no-inline-config)", "result": "4049 files, 0 errors, 11581 pre-existing warnings. The 8 warnings on touched files are pre-existing no-explicit-any on lines this diff does not touch." },
 { "name": "check:control-bytes", "command": "pnpm run check:control-bytes", "result": "exit 0" },
 { "name": "check:phantom-deps", "command": "pnpm run check:phantom-deps", "result": "exit 0 -- covers the two new devDependencies edges" },
 { "name": "check:spec-symbols", "command": "pnpm run check:spec-symbols", "result": "exit 0" },
 { "name": "check:vi-mock-specifiers / check:vi-mock-inherit", "command": "pnpm run check:vi-mock-*", "result": "exit 0" },
 { "name": "check:action-forward-parity / check:designer-field-key-parity", "command": "pnpm run ...", "result": "exit 0 -- both read enum vocabularies, so both are downstream of this change" },
 { "name": "check:readme-exports", "command": "pnpm run check:readme-exports", "result": "GREEN: the gate's own verdict line reads 'check-readme-exports: OK (43 tracked README(s) ... 386 self-imports judged (386 real, 0 wrong-path, 0 fabricated))'. Its FIRST run exited 1 naming 'type entry ./dist/index.d.ts is not on disk -- run pnpm build first' for 7 packages outside this diff; building them turned it green. Recorded as a build PREREQUISITE, not as a failed measurement." }
 ],
 "claim_comment_url": "https://github.com/objectstack-ai/objectui/issues/7025#issuecomment-5480329032",
 "assignees_untouched": true,
 "github_api_ok": false,
 "github_api_note": "PARTIAL, and reported rather than worked around. MCP reads and WRITES all succeeded (issue reads, claim comment, PR create, PR update, this comment). ONE call failed: mcp__github__search_issues returned 'API rate limit already exceeded for user ID 318158314' -- the search/GraphQL quota, the same limit the dispatch says the PM seat is under. Nothing was skipped because of it: the required dedupe read for the out-of-scope finding was completed over the anonymous repo-scoped REST channel instead (probe first: GET /repos/objectstack-ai/objectui returned HTTP 200; global /search/issues is blocked by the session proxy with 'sessions are bound to their configured repositories', so the dedupe used the repo-scoped label-filtered list endpoint plus a local grep, then a direct read of #5872). CHANNEL SWITCH DECLARED. No blanket-table scan was run on either channel.",
 "mcp_calls": 9,
 "open_questions": [],
 "blockers": []
}

---

Generated by Claude Code