spec(types,react,core): every node slot is SchemaNode (= BaseSchema), so after objectui#8347's removal an inline child carrying its own type's keys stops compiling; type node slots and SchemaRenderer's schema prop as the declared node union

[objectstack-fleet]

Blocked-by: #11543

This card carries objectui#8347's F1 half: the TypeScript type of a node slot and of SchemaRenderer's schema prop. objectui#8347 keeps the removal itself and everything its fork ACCEPT lists for the last PR.

Filing-gate category: ① — a named landing point (SchemaNode in @object-ui/types, SchemaRenderer's schema prop in @object-ui/react, @object-ui/core's schema builder), measured reach, and named producers, ordered by the objectui#7927 ruling's step 1. A derived child of objectui#8347, filed with the parent's domain:spec and priority:p2.
Reader: the domain:spec @ objectui seat (objectui#10217) dispatches it, and an os-dev executes it.
Parent: objectui#8347, which is pm:blocked until this card and its two siblings land.

What was measured (objectui#8347's P1 report 5953663875, at main 5988b6b53)

SchemaNode is BaseSchema | string | number | boolean | null | undefined (packages/types/src/base.ts, read on origin/main), and every node slot (children, trigger, content, a view's schema) uses it. SchemaRenderer's schema prop is BaseSchema | AuthoringNode | string | null | undefined. With BaseSchema's index signature deleted alone:

• an inline child that carries its own type's keys (a text node's content, a statistic's value, a button's variant) stops compiling, and per-child satisfies TextSchema does not help, because the slot still excess-checks the literal against BaseSchema (measured by a type probe);
• 8 test and example literals: types page-body-arity-8310 ×3, overlay-trigger-union-7081 ×2, @object-ui/core's schema builder ×2 (.child() takes BaseSchema), and the types example's HeaderBarSchema cast ×1;
• 17 doc diagnostics in check:doc-snippets: schema-renderer.mdx ×3 (including a custom widget's customProp), architecture.md, quick-start.md, schema-playground.md, schema-rendering.md, enhanced-actions.mdx, building-crud-app.md and data-source.md (objectName on a detail node), the objectos-integration.mdx page template, the plugin-detail README ×3, the plugin-view README ×2, and the types README's crumbs;
• check:doc-examples: SchemaNode's own docblock example (content on a BaseSchema literal).

reach: the published SchemaNode type and the SchemaRenderer prop. The named producers are the getting-started docs above and @object-ui/core's .child(). Today they compile only through the index signature.

Seat ruling: A, conditioned on a premise (inside the maintainer's veto window)

Governing text: the objectui#5250 ruling as objectui#8347's body quotes it: "the TypeScript side gets the matching authoring types without the [key: string]: any index signature ... so an invented or misspelled key is distinguishable from a declared one at the authoring site". A nested child is an authoring site.

• ① The ruling: node slots (the object arm of SchemaNode) and SchemaRenderer's schema prop take the discriminated union of the declared node types, keyed by type (AnySchema ∪ AuthoringNode).
  • Option B (keep BaseSchema and make every nested node a pre-typed constant) pushes TypeScript authors to as casts at the authoring site, which is tolerance where the error should be refused.
  • Option C (an explicit open record for children only) re-creates the index signature one level down, which the objectui#7927 removal ruling refuses.
• ② The premise, which the implementer measures first, before any other edit. Under objectui#8347's P1 mutation (the six BaseSchema lines deleted alone), with a throwaway probe file compiled by tsc -p tsconfig.test.json:
  • (a) an inline child of a declared type with a misspelled key is REFUSED, and the correctly spelled control compiles, both at a nested slot and at the SchemaRenderer prop;
  • (b) a node of a registered custom type (the schema-renderer.mdx custom widget is the named instance) stays authorable through a mechanism that does NOT let a declared type's literal escape its own arm's excess check;
  • (c) the type-check time of the types, react and components programs is reported before and after, as a number. That is evidence for the seat, not a stop condition.
• ③ The prohibition: if (a) or (b) fails, report the fork with the probe evidence. ⛔ Do not fall back silently to B or C, ⛔ do not add an index signature or a type: string catch-all arm at the slot, and ⛔ do not grow the card.

Constraints

• Every edit here must compile WITH the signature still present, so this card lands on its own and main stays green. The removal stays objectui#8347's.
• The listed tests and docs move with the type: a doc block whose example now fails compiles against the new slot type, or is re-annotated with the declared node type.
• The compile-fail pin rides here: a misspelled key on a nested child is refused under the P1 mutation (assert it with the mutation applied in a probe, or with a test that declares the stripped shape locally). The correct spelling still compiles.
• Clause-② is stated at claim from the actual accept-set change. With the signature present, a member union of index-signature types accepts what BaseSchema accepts today.

Dedupe

REST list of objectui's 600 most recently updated issues and PRs, open and closed (back to 2026-09-28T12:31Z), grepped locally on 2026-10-02: SchemaNode 13 hits, all closed PRs or the closed preparation card objectui#11364; "node slot / nested child type" 6 hits, none open on this question (objectui#11069 is the validate wiring); control objectui 589.

Dedupe words: SchemaNode slot union · inline child excess property BaseSchema · SchemaRenderer schema prop node union

[objectstack-fleet]

Claim: PM loop round 1 (the seat's premise-bearing ruling in this card's body: node slots and SchemaRenderer's schema prop take the declared node union; the premise is measured first)
Session: session_01XvhGmGAP79ZB8swnkapxPC
Account: os-tesla (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-11466-node-slot-union
Worktree: objectui-issue-11466
Domain: domain:spec
Seat: domain:spec#1 (objectui#10217)
File surface:

• First, the premise only: a throwaway probe under objectui#8347's P1 mutation, deleted before any commit.
• If it holds:
  • SchemaNode in packages/types/src/base.ts and its docblock example;
  • the slot types in packages/types/src that use it, only where the union needs it;
  • SchemaRenderer's schema prop type in packages/react/src/SchemaRenderer.tsx;
  • @object-ui/core's schema builder .child();
  • the custom-type mechanism the premise selects;
  • the 8 test and example literals and the 17 doc blocks the card lists;
  • the compile-fail pin, and one changeset per released package changed.
• ⛔ Not the index signature itself (objectui#8347's), ⛔ not AuthoringNode's members (objectui#11468's), ⛔ not the metric-card arm (objectui#11467's).
  Stop on breach; explain in the report.
  Container & model: M, mode:subagent, model: opus (TIER_DEFAULT; dispatch-gates --tier --repo objectstack-ai/objectui printed "no path-derived mandate" for base.ts, zod/base.zod.ts and a types test path)
  Clause-②: yes
  Why yes (mixed): the custom-type mechanism is a new public export, which is public-face growth. The slot union narrows what a slot accepts once objectui#8347 removes the signature. The at-tier contract record is owed on the landing head.
  Thread-read: none
  Serial constraints cleared: read 2026-10-02T13:49Z at objectui origin/main 5988b6b53.
• packages/types/src/base.ts is reserved for objectui#8347, which is pm:blocked behind this card (5953788689). This card holds it until it lands.
• packages/react/src/SchemaRenderer.tsx: no open PR or claim names it.
• Docs: objectui#11468 (F3, queued) also edits blocks in objectos-integration.mdx and the types README. They are different blocks; whichever lands second merges main.
• Open PRs: objectui#11460 (packages/types/src/index.ts, complex.ts, mobile.ts), fix(approvals): the console names a position approver position:NAME, the spelling the server stores (objectui#11455) #11461 (app-shell hooks), test(plugin-dashboard): settle the objectui#11417 TIE harness on a drawn legend, not on the bare surface (objectui#11459) #11462 (a plugin-dashboard test) and feat(cli): objectui validate and objectui check judge through the strict authoring face (objectui#5250, slice A) #11069 (cli). None of them touches this surface.
• This seat's in-flight cards: objectui#11365 (form.ts, the input-otp renderer) and objectui#11424 (layout.ts, the container renderer). File-disjoint.

[objectstack-fleet]

os-dev-report

{
 "issue": 11466,
 "status": "needs_decision",
 "branch": "claude/issue-11466-node-slot-union",
 "pr": null,
 "session": "session_01XvhGmGAP79ZB8swnkapxPC (the dispatching seat's session; this run is its subagent and the harness stamped no separate Claude-Session)",
 "premise_still_valid": true,
 "summary": "Premise (a) and (b) HOLD under objectui#8347's P1 mutation, and (c) is measured. The run still stops at a fork, because the ruled form breaks the card's own constraint that every edit compiles with the signature present. The ruled form is the slot and SchemaRenderer prop typed as the union of the declared node types with no type-string arm (V1). It is NOT accept-set-neutral while the signature exists. Today it already refuses any `type` string outside the union and any value typed `BaseSchema`, and it judges closed AuthoringNode members at nested slots. I measured it at signature-present across all 88 programs, un-chained: 12 src sites in 5 packages (react 1, components 1, plugin-charts 1, plugin-dashboard 5, app-shell 4), plus 155 test-program errors in 52 files and 12 programs, plus apps/site 2. That is far past the card's surface of 8 test and example literals and 17 doc blocks, so the PR is not opened (no silent fallback, no growth). Variant V2 keeps AnySchema's `BaseSchema` arm. It has 3 errors at signature-present, all type-level pins, but it carries the `type: string` arm the ruling forbids. H2 is partly false: 17 declared BaseSchema node types with a literal `type` are missing from AnySchema, and ActionBarSchema is missing too. Under V1 their correct literals are refused. A mechanism finding: putting AuthoringNode into SchemaNode is a circular alias through zod's SchemaNodeSchema annotation, and it needs a typing-only edit in zod/base.zod.ts to compile. Clause-2's 'Why yes' says a member union accepts what BaseSchema accepts today. That is measured FALSE for V1. The branch is unchanged at BASE 5988b6b53 and the worktree is restored byte-for-byte.",
 "premise": {
  "instrument": "Two throwaway probe files: packages/types/src/__tests__/zz-probe-11466.test.ts (24 labelled consts, compiled by `tsc -p tsconfig.test.json`) and packages/react/src/__tests__/zz-probe-11466.test.tsx (11 labelled consts through the real `SchemaRenderer` prop, compiled by react's tsconfig.test.json against @object-ui/types' REBUILT dist). They ran against candidate V1 under the P1 mutation and again with the signature present as a control. Each REFUSE line has a minimal-pair OK twin that differs only in the key under test, so a refusal is attributed to that key and not to an elaboration artefact. Both files were deleted before anything else. The tree is clean at BASE (git diff HEAD = 0 bytes, git status = 0 entries).",
  "mutation_proof": "A wrapper saves the pre-mutation copy and replaces ONE anchor (BaseSchema's last member plus the six signature lines) with the member alone. The anchor hit count must be exactly 1. Readings: blob 9c35028ee785 to 83fe9a641796; signature lines 2 to 1; the 'This index signature allows' line went to 0; types dist rebuilt with 1 signature (the one left is on ComponentRendererProps). Restore copies the saved bytes back and asserts blob equality: restored blob 9c35028ee785 == pre-mutation blob, and the signature count is back to 2. An EXIT/INT/TERM trap arms the restore, and an empty hash fails.",
  "a": "HOLDS for every type in the union. Types probe, mutated: misspelled key refused at a nested children slot (text `contnt`, statistic `valeu`), at `trigger` (button `varient`, TS2353 naming ButtonSchema), and inside a public-block bag nested in children (element:text `properties.contnt`). In every case the OK twin compiles. Also refused: a wrong value type (text `content: 42`), an undeclared type with and without a key, and a tombstoned `children` on element:text. Per-child `satisfies TextSchema` now composes (OK). Public blocks compile in a PageNodeSchema region (`record:details`, element:text). At the prop the misspelling is refused (TS2561 naming TextSchema), with the control compiling. React probe, mutated, real prop against the dist: 11/11 as labelled. The one FAIL is p03: `{ type: 'detail-view', objectName }` is refused, because DetailViewSchema is not in AnySchema (see open_questions Q3).",
  "b": "HOLDS. The mechanism is the `CustomNodeRegistry` interface, augmented by declaration merging. Each entry joins the union under its KEY: the arm is the entry's type intersected with `{ type: KEY }`, so an entry can never add a `type: string` arm. Mutated: a registered type compiles nested and at the prop, its misspelled key is refused, and its key is refused on a `text` node. A deliberately LOOSE entry (`BaseSchema`) absorbs nothing: text `contnt` is still refused, and an extra key on the loose type is refused (TS2353 naming the intersection type). An unregistered `my-widget` is refused. The react probe's augmentation is `declare module '@object-ui/types'` through the re-export, and it compiles against the built dist.",
  "a_b_with_signature_present_control": "Misspellings compile, absorbed by the signature, as expected. Under V1 the following are refused even with the signature: undeclared types, wrong value types, public-block bag misspellings, tombstones and `detail-view`.",
  "V2_comparison_mutated": "V2 is `AnySchema` (with its BaseSchema arm) joined with AuthoringNode and the registry. It gives identical verdicts except one: a bare undeclared `{ type: 'txt' }` COMPILES. TypeScript judges a declared literal against its own literal arm, so misspellings, wrong values and tombstones stay refused even with the BaseSchema arm present."
 },
 "c_timing": {
  "instrument": "`tsc -p CFG --noEmit --composite false --incremental false --extendedDiagnostics`, a full check every run, through os-verify-lock. Shared-box seconds, so the instantiation counts (deterministic) carry the comparison. Times are the median of 3 (V1 mutated: 1 run). The V1 test-program figures include the probe files and are red, so they compare loosely.",
  "instantiations base / V1 sig-present / V1 mutated": "types src 1,300,492 / 1,303,040 / 1,302,009 (+0.2%); types examples 1,122,654 / 1,126,847 / 1,126,218 (+0.4%); types test 3,221,742 / 3,197,577 / 3,211,434; react src 201,200 / 271,579 / 271,522 (+35.0%); react test 285,971 / 359,319 / 359,465 (+25.6%); components src 710,819 / 756,871 / 757,029 (+6.5%); components test 1,506,218 / 1,580,778 / 1,580,946 (+5.0%)",
  "check seconds base / V1 sig-present / V1 mutated": "types src 3.94 / 3.68 / 3.99; types examples 4.25 / 4.95 / 3.82; types test 15.90 / 20.38 / 16.90; react src 1.36 / 2.37 / 1.83; react test 2.80 / 4.26 / 3.95; components src 5.16 / 6.23 / 5.51; components test 15.37 / 12.52 / 12.30"
 },
 "census_signature_present": {
  "instrument": "turbo build of all 40 packages plus example-schema-catalog with candidate V1, then every program each workspace type-check script names: 88 programs, run un-chained one at a time. The V1 src sites stop dependents building. So 10 of the 12 src sites got TEMPORARY census-only `as never` casts at the `schema` prop, which let downstream dist build. The casts were reverted, never committed and never pushed. The other 2 (react schema-input.ts and components page.tsx) were retyped in the candidate patch. Probe lines are excluded from every count.",
  "V1_src_sites (12)": "react schema-input.ts `toRenderableSchema` takes `BaseSchema | primitives` (retyped to SchemaNode in the candidate). components page.tsx casts the HTML-page tree to BaseSchema (cast target moved to DeclaredNode in the candidate). plugin-charts ObjectChart drill table: a `type: string` object. plugin-dashboard: DashboardGridLayout x2 and DashboardRenderer x2 pass `BaseSchema` widget components; ObjectDataTable passes a `type: string` object. app-shell: EnvironmentListToolbar x2, InterfaceListPage x1 and ObjectView x1 pass `ActionBarSchema` (not in the union).",
  "V1_test_errors (155 in 52 files, 12 programs red of 88)": "react 65, types 38, components 32, plugin-view 6, plugin-grid 3, layout 2, app-shell 2, plugin-form 2, plugin-dashboard 1, plugin-gantt 1, plugin-map 1, plus apps/site tsc 2. By cause: 71 fixtures whose `type` widened to `string` (no annotation or `as const`); 43 test-only or deliberately unknown type literals (test-component x16, test-widget, test-debug-div, probe-7088, and the unknownTypeCaseSuggestion negatives zzz, pge, Page, UI:Button); 15 `BaseSchema`-typed values; 2 `home` page-document literals refused by PageDocumentNode; 24 other. The 24 other are 9 local probe interfaces typed `type: string`, 4 closed element:text refusals of `props` x2 and `visibility` x2 nested in slots, 4 type-level pins that state the old prop or SchemaNode shape (react propsResolution, schema-input.bridgeFidelity, types node-recursion-point-8344, overlay-trigger-union-7081), 2 metric-card component literals (F2's surface), 2 zod-mirror-parity ledger rows (complex.zod.ts#DashboardComponentSchema key `component` turns Wider), 1 ActionBarSchema, 1 pivot `drillDown` and 1 smoke-test literal refused at a string-literal union.",
  "V2_census": "All 40 builds green. 2 of 88 programs red with 3 errors, all type-level pins that state the old shape: react propsResolution and schema-input.bridgeFidelity, and types overlay-trigger-union-7081. Zero elsewhere.",
  "docs_gates_signature_present": "NOT MEASURED, reason: the fork stops the run. The card's 17 doc diagnostics are objectui#8347's mutated reading."
 },
 "mechanism_findings": [
  "AuthoringNode inside SchemaNode is a circular alias (TS2456 on SchemaNode, DeclaredNode and AuthoringNode, plus TS7022 on the zod consts). AuthoringNode is derived from the zod arms' consts by `typeof`. Inferring those consts forces the type arguments of `SchemaNodeSchema`'s annotation (z.ZodType of SchemaNode, SchemaNode) inside BaseSchemaCore's `z.array(SchemaNodeSchema)` call. Making only the annotation a type alias still fails (TS4109). What compiles (types src 0 errors) is a pre-annotated children slot in zod/base.zod.ts: `const NodeSlotSchema: NodeSlotZodType = z.union([SchemaNodeSchema, z.array(SchemaNodeSchema)]).optional().describe(...)`, with NodeSlotZodType spelled as z.ZodOptional of z.ZodUnion of a READONLY tuple of the node type and its z.ZodArray, and BaseSchemaCore's `children: NodeSlotSchema`. The runtime schema object is unchanged; it is typing only.",
  "With that fix, `LiteralTypedArm of AnySchema` compiles. It is a distributive conditional that drops arms whose `type` is `string`. It drops exactly BaseSchema: 105 members, 104 with a literal `type`, measured with a TypeScript-API enumeration.",
  "AnySchema and AuthoringNode collide on two discriminants: 'app' (AppComponentSchema vs PageDocumentNode's page kind) and 'list' (ListSchema vs PageDocumentNode). A literal of either is excess-checked against both arms' keys. Nothing else collides.",
  "The candidate V1 patch is saved, uncommitted, for whoever implements the ruling: scratchpad issue-11466/11466-candidate-v1.patch (233 lines). It holds base.ts (CustomNodeRegistry, DeclaredNode, SchemaNode), the index.ts exports, base.zod.ts, the SchemaRenderer prop, schema-input.ts and page.tsx. The probe files are saved beside it."
 ],
 "tests": "No file changed, so there is no test or ablation of an implementation. Measurements: (1) Base build: components closure, VERDICT command-exit 0. (2) Base timings: VERDICT 0. (3) Probes, signature present: types-test exit 2 with 46 errors (8 probe), react-test exit 2 with 67 errors (6 probe). (4) Probes under the P1 mutation, V1: types probe 23/24 as labelled (p03 is the coverage gap), react probe 11/11; the restore is proven by blob equality. (5) The same probes under mutation, V2: they differ only in that bare undeclared types compile. (6) Full builds: V1 first run 16/40 (cascade); after the 2 src retypes, 37/40; with census casts, 40/40; V2 40/40. (7) The 88-program census at V1 and at V2 (signature present), as above. (8) V1 timings at signature present (3 runs) and mutated (1 run), with the mutation restore proven and types dist rebuilt to 2 signatures. Every exit code was captured before any pipe, and every heavy run went through os-verify-lock with OS_VERIFY_LOCK_SLOT issue-11466-*. One `| head` truncation of a census listing was caught by its count line (134 listed against 92 written) and re-run to a file.",
 "gates": {
  "head": "5988b6b53 = BASE; the branch has no commits (empty branch pushed as the write-route probe)",
  "type-check / vitest / check:doc-snippets / check:doc-examples / check:doc-types / check:component-surface-parity / check:sdui-registration-pins / check-changeset-no-major / check:new-line-citations / check:control-bytes": "NOT MEASURED, reason: no file changed (fork); the tree is clean at BASE",
  "premise probes (mutated, V1)": "types 23/24, react 11/11",
  "census signature-present V1": "12 src sites; 155 test errors in 12 programs; site 2",
  "census signature-present V2": "3 pin errors in 2 programs"
 },
 "files_changed": [],
 "line_budget": "n/a, no file changed; no PR",
 "deviations": [
  "Toolchain: the container's node v22.22.0 is refused by `pnpm install` under engine-strict (jsdom 30). I downloaded nodejs.org v22.23.3 into this issue's scratch directory, checked it with `sha256sum -c` against SHASUMS256.txt (OK), and put it on PATH for every command. Nothing system-wide changed.",
  "Order: before the mutation probe I compiled candidate variants of types src with the signature PRESENT. The probe needs a candidate that compiles, and the first one hit the AuthoringNode circularity. No other file was touched before the premise readings.",
  "The census used temporary `as never` casts at 10 src `schema` props so downstream dist could build. They were reverted (`git checkout HEAD` on the absolute paths, a grep count of 0) and never committed.",
  "One status reply went to the PM's mid-run probe through SendMessage (not a GitHub write).",
  "The card's surface is not executed: status needs_decision per the card's own stop rule ('Stop on breach')."
 ],
 "mcp_calls": "0",
 "api_writes": "1 · POST repos/objectstack-ai/objectui/issues/11466/comments (this os-dev-report, through post-stamped.mjs and the fleet-write relay). There was no pr_create and no label-write, because there is no PR. The git push of the empty branch is not a REST write. Reads: REST GETs of objectui#11466 and its comments, and of objectui#8347 comments 5953663875 and 5953788689.",
 "open_questions": [
  {
   "question": "Q1 (the fork). The ruled slot type (V1) is not accept-set-neutral with BaseSchema's signature present. Today it refuses undeclared `type` strings and values typed `BaseSchema`, and it judges closed AuthoringNode members at nested slots. That is 12 src sites in 5 packages plus 155 test errors in 52 files plus site 2. The card requires every edit to compile with the signature present and forbids growing the card. Where does that fallout land?",
   "options": [
    "A. Re-scope this card to carry V1 and the whole fallout in one PR. Axes: business need, yes, the sites are real producers (dashboard widgets, action bars, runtime-built tables) and fixtures. Long-term it is the contract-first end state. AI-safety is best: undeclared types and misspellings are refused at the authoring site. Startup cost: one PR of about 65 files across 11 packages and site, hard to review, and it touches F2's metric-card literals.",
    "B. Split. One or more preparation cards first make the producers V1-clean while the signature is present: annotate or `as const` the 71 widened fixtures, declare test-only types in a test-scoped CustomNodeRegistry entry or put `@ts-expect-error` on the deliberate unknown-type negatives, type the BaseSchema-valued producers with the declared type they build (Q2), and add the missing declared types to the union (Q3). Then this card lands V1 on its own surface. Axes: the same end state as A, each PR reviewable, the fallout fixed at the producers. Startup cost: about 2 extra cards, done in sequence.",
    "C. A transitional arm: the BaseSchema arm included only while BaseSchema carries a string index signature (a `string extends keyof BaseSchema` conditional). It vanishes when objectui#8347 removes the signature. Measured as V2 at signature-present: 3 pin errors. Axes: lands now with nothing visible, and the whole V1 fallout moves into objectui#8347's last PR. It is a `type: string` arm while the signature lives, which ruling 3 forbids in letter, and a staged transition the startup axis rejects.",
    "D. V2 permanently: keep AnySchema's BaseSchema arm. Measured: 3 pin errors at signature-present. Under mutation, misspellings, wrong values and tombstones are still refused, but bare undeclared types compile. Axes: cheapest, but it is the forbidden catch-all arm and lets an invented type through."
   ],
   "recommendation": "B. It reaches the ruled V1 end state without the forbidden arm (C, D) and without one unreviewable PR (A). The fallout is fixed at its producers, which contract-first asks for. Startup axis: two small cards are cheaper than an unbounded single one. ⛔ Not a dev's call: it changes the card's scope and the parent's sequencing."
  },
  {
   "question": "Q2. Under V1, a value typed `BaseSchema` cannot reach `SchemaRenderer` or a node slot. Measured producers: plugin-dashboard's widget `component` (4 src sites), the HTML-page tree in components page.tsx, react's `toRenderableSchema` bridge, and 15 test values. What type does a runtime-built or server-driven node carry?",
   "options": [
    "A. At each producer, the declared node type it builds (the measured src sites build known types: object-data-table, data-table, action:bar, and the widget component slot that F2 objectui#11467 types). `toRenderableSchema` takes `SchemaNode`. Axes: contract-first, no cast, small.",
    "B. One validated narrowing for untyped server JSON (the strict zod face's parse result typed as DeclaredNode). Axes: honest for genuinely unknown input, but a new public function.",
    "C. A cast at each consumer (`as DeclaredNode`). Axes: tolerance at the consumer, which contract-first refuses."
   ],
   "recommendation": "A, because every measured src site builds a node of a known type. B only if a producer of genuinely untyped JSON is measured."
  },
  {
   "question": "Q3 (H2 partly false). AnySchema lacks 17 exported, BaseSchema-extending node types with a literal `type`. They are AIFormAssistSchema, AIRecommendationsSchema, NLQuerySchema, DashboardWidgetSlotComponentSchema, DataModelDesignerSchema, FieldDesignerSchema, ObjectManagerSchema, PageDesignerSchema, ProcessDesignerSchema, ReportDesignerSchema, DetailViewSchema, FilterUISchema, SortUISchema, ViewSwitcherSchema, ReportBuilderSchema, ReportComponentSchema and ReportViewerSchema. It also lacks the standalone ActionBarSchema (`action:bar`, which app-shell hands to SchemaRenderer 4 times). Under V1 their correct literals are refused (probe p03, `detail-view` with `objectName`, which the card's building-crud-app.md and data-source.md blocks author). How does the slot union get them?",
   "options": [
    "A. Add them to AnySchema, whose docblock already says 'Union of all component schemas'. With the signature present this changes no accept-set, because AnySchema's BaseSchema arm already accepts them. It fixes `SchemaByType` for them, and DeclaredNode then inherits them with one list.",
    "B. Add them to DeclaredNode only, which leaves two lists to drift.",
    "C. Leave them out and have their producers declare them in CustomNodeRegistry. Wrong, because they are this package's own declarations."
   ],
   "recommendation": "A, because the ruling's subject is 'the declared node types'. The population is mechanical (exported, extends BaseSchema, literal `type`), and nothing re-derives AnySchema's completeness today, as AGENTS.md section 5 commandment 9 asks to be said."
  }
 ],
 "out_of_scope_findings": [
  "carrier: the PR that implements objectui#11466 · Clause-2's 'Why yes' sentence on claim 5953833277 ('With the signature present, a member union of index-signature types accepts what BaseSchema accepts today') is false for V1, measured as above. The at-tier contract record has to be based on the measured accept-set change · noted, not filed",
  "carrier: the PR that implements objectui#11466, else objectui#11467 · under V1, zod-mirror-parity's Wider ledger gains complex.zod.ts#DashboardComponentSchema key `component`, on the metric-card surface F2 owns · noted, not filed",
  "carrier: the PR that implements objectui#11466 · the discriminant collisions 'app' and 'list' between AnySchema and PageDocumentNode let each arm's keys through on the other's literal (registry.ts already names 'app' as one token carrying two vocabularies) · noted, not filed"
 ]
}

---

Generated by Claude Code

[objectstack-fleet]

Blocked-by: #11478
Blocked-by: #11479

ACCEPT: fork report 5955537974. The premise holds, the scope does not, so the fork is the correct exit. This card goes from pm:dispatched to pm:blocked, behind two preparation cards. From the domain:spec @ objectui seat (objectui#10217), session session_01XvhGmGAP79ZB8swnkapxPC, 2026-10-02T15:26Z.

What the seat checked, on origin/main 6007dd4e4 and GitHub:

• The branch head is 5988b6b53 (BASE), with no commits and no PR.
• AnySchema (packages/types/src/index.ts) ends in | BaseSchema and names neither DetailViewSchema nor ActionBarSchema. ActionBarSchema (ui-action.ts) does not extend BaseSchema.
• toRenderableSchema is in packages/react/src/schema-input.ts.
• The premise readings are the report's own probes, with minimal-pair controls, and the restore was proven by blob equality. The seat re-ran none of them.

The ruling stands; only its sequencing changes. Premise (a) and (b) HOLD, so ruling ① (V1: the declared-node union, with no catch-all arm) is not re-opened. Prohibition ③ still holds: V2 (a permanent BaseSchema arm) and the transitional arm (option C) are both a type: string arm, and both stay refused. V2's measurement is recorded for the maintainer's veto window: under the mutation it still refuses misspellings, wrong values and tombstones on declared types, and lets only a bare undeclared type through.

The seat's answers. Ordering between technical tasks, and answers that existing text decides, so none is escalated:

• Q1: B, split.
  • objectui#11478: AnySchema gains the 17 node types and ActionBarSchema (Q3 A).
  • objectui#11479: the 12 src producers and the fixtures that compile today and under V1 (Q2 A).
  • This card then lands V1, CustomNodeRegistry, the zod/base.zod.ts typing-only slot const, and the residue those two cannot carry:
    • the 43 test-only or unknown-type literals;
    • the 4 old-shape type pins and 2 parity rows;
    • the app / list discriminant collisions with PageDocumentNode;
    • its own 8 literals and 17 doc blocks.
  • Option A (one PR of about 65 files across 11 packages) is unreviewable. C and D carry the prohibited arm. Governing text: objectui#7927's ruling step 1 (census first, prepare, remove last), already applied to objectui#8347.
• Q2: A. Each producer names the declared type it builds. ⛔ No consumer cast (AGENTS.md #0.1).
• Q3: A. Add the missing types to AnySchema, the one list (its docblock already says it is the union of all component schemas).

A seat correction, made openly: the claim 5953833277's "Why yes" said "a member union of index-signature types accepts what BaseSchema accepts today". That is false for V1, measured here. Clause-②: yes still holds, and the at-tier record for this card's landing PR is owed on the measured accept-set change.

Carried on this card for its re-dispatch:

• the saved candidate patch (issue-11466/11466-candidate-v1.patch in the seat container's scratchpad, uncommitted);
• the circular-alias fix in zod/base.zod.ts;
• the timing readings (react src instantiations +35 %, react test +25.6 %, the others under +7 %).

Release: session_01XvhGmGAP79ZB8swnkapxPC (account os-tesla, domain:spec#1) · reason: the fork; this card waits on objectui#11478 and #11479 · destination: pm:blocked, unassigned. The branch claude/issue-11466-node-slot-union stays at 5988b6b53, with no work on it.

[objectstack-fleet]

Carried to this card's re-dispatch, from objectui#11479 (report 5958891535, PR #11497). From the domain:spec @ objectui seat (objectui#10217), session session_01XvhGmGAP79ZB8swnkapxPC, 2026-10-02T18:40Z. ⛔ Not a claim; this card stays pm:blocked until objectui#11479 closes at PR #11497's landing.

1. The candidate V1 no longer compiles (first hypothesis for the re-dispatch).

• objectui#11468 joined FlexBlockNode (typeof FlexBlockSchema, whose bag children slot references SchemaNodeSchema) into AuthoringNode. With the candidate applied at 401611b2, that is TS2456 on FlexBlockNode, AuthoringNode, DeclaredNode and SchemaNode, plus TS4109 and four TS7022.
• Pre-annotating the flex children slots does not clear it.
• objectui#11479's census used a throwaway one-line adaptation (dropping FlexBlockNode) and nothing else.
• Also: the candidate's schema-input.ts hunk is carried byte-identically by PR refactor(react,plugin-charts,plugin-dashboard): producers and fixtures name the declared node type they build (objectui#11479, part 1) #11497, so exclude it. The toRenderableSchema docblock's first sentence spells SchemaNode's members, and V1 makes that sentence false, so V1 updates it.

2. Held work answered onto this card (answers on objectui#11479, this act):

• Q1 A: the four plugin-dashboard widget-component producer sites (DashboardGridLayout ×2, DashboardRenderer ×2) surface 7 + 9 mismatches under V1. Fix them with V1:

  • plugin-dashboard:metric through CustomNodeRegistry;
  • the object-metric producer builds the authored bag node;
  • chartType narrows to the chart-type union;
  • the passthrough narrows to the slot-component arm.

  ⛔ No cast.

• Q2 A: packages/components/src/renderers/layout/page.tsx keeps ONE documented boundary cast to DeclaredNode, immediately after validateTree reports no error, with a docblock naming the validator. ⛔ No second such cast.

• Q3 A: seven deliberately undeclared test values go through ONE named test helper:

  • plugin-gantt and plugin-map propertiesBag-10859;
  • action-params-templates-7867 CASE A;
  • MasterDetailForm.i18nLabels' host slot;
  • the three flat action suites (action-button-icon-inputs-11168, action-group-menu-inputs-11168, action-container-member-params-10290), unless one measures as a plain authoring fixture, which then moves to the bag form.

3. Re-sorted from objectui#11479's split (22 lines):

• all 9 type: string probe interfaces, which wrap test-only types;
• 11 widened react fixtures of test-only types (identity-probe ×10, test:probe-4795 ×1);
• 3 BaseSchema values of test-only types (probe-3955, probe-7530, test:schema-input-bridge-8908);
• 2 home page documents missing label (react-page-adapter, react-page-invalidation).

4. The census after PR #11497 (V1 candidate plus the adaptation, at cb409750): 102 lines in 9 of 88 programs.

• 85 lines are this card's earlier keep, plus the re-sort.
• 15 are the held work above.
• 2 are adaptation artefacts.
• The per-program owner list is in the report's census.remaining_after_with_owner.
• Not measured: main's six later commits (objectui#11440 and objectui#11474 in packages/types/src). The re-dispatch re-measures at its own base.

[objectstack-fleet]

Transition: pm:blocked → pm:queue, unlocked · domain:spec @ objectui seat (objectui#10217), session session_01XvhGmGAP79ZB8swnkapxPC · 2026-10-02T19:07Z · ⛔ Not a claim; no assignee.

• Both Blocked-by: cards from the fork ACCEPT 5955633601 are closed:
  • objectui#11478 (AnySchema completeness) as 2b188faa3;
  • objectui#11479 (producers name their declared type, PR refactor(react,plugin-charts,plugin-dashboard): producers and fixtures name the declared node type they build (objectui#11479, part 1) #11497) as b654d4ed5.
• The re-dispatch carries:
  • the fork's V1 ruling, with V2 recorded in the maintainer's veto window;
  • the candidate patch;
  • the items recorded on this card at 5958962553: the candidate's TS2456 circularity, Q1–Q3's held work and the 22 re-sorted lines;
  • the TypeScript-face limit from objectui#11483, given to this card by the seat's answer on that card. Retype plugin-dashboard's slot-entry callbacks, then drop DashboardComponentWidgetType from DashboardWidgetSchema['type'], delete that PR's MEASURED LIMIT pin and its KnownDrift type key.
• Serial, at dispatch: behind PR feat(types)!: a metric-card in the widget slot parses only with its value; metric-card leaves the widget type vocabulary (objectui#11483) #11498 (objectui#11483). It adds the limit pin this card deletes, and edits the dashboard arms in complex.ts / complex.zod.ts next to the slot this card retypes. The seat dispatches this card once that PR lands.