finding(components): eleven more `SchemaNode` slots carry the same numeric-falsy JSX trap objectui#9033 closed on `header-bar` — the census the card's grep could not see

[os-tesla]

Observation-class finding, measured while implementing objectui#9033. Filed unassigned, no labels — grading is the triage seat's. Generated by Claude Code, session session_01UzHd6hDYatoDn17BuwKxnZ, working the objectui#9033 dev seat.

The claim

objectui#9033 said of header-bar's rightContent: "It is the only remaining instance", on the strength of git grep -n '&& .*toRenderableSchema' returning exactly one line. That grep is scoped to the toRenderableSchema spelling of the right operand. The trap does not depend on that spelling — it depends only on the LEFT operand being a value React renders. Every SchemaNode-typed slot guarded by a bare && has it, whatever sits on the right.

Re-run with the spelling constraint dropped, eleven more sites leak, all in @object-ui/components, all published renderers. objectui#9033's own site is fixed; these are not.

Measured, on origin/main at 3b6bc6959

A temporary probe rendered each slot through the real renderer stack and read document.body.textContent (Radix overlays portal out of the render container, so reading the container alone is a blind instrument — the first pass of this probe read three overlay rows as "clean" for exactly that reason). Each row carries its own lit control: the same slot given 42, which must reach the text or the row reports NOT MEASURED instead of clean. The probe was deleted afterwards; git status --porcelain empty.

renderer   slot      verdict
container  children  LEAK    baseline=[] with-0=[0]
flex       children  LEAK    baseline=[] with-0=[0]
grid       children  LEAK    baseline=[] with-0=[0]
stack      children  LEAK    baseline=[] with-0=[0]
card       header    LEAK    baseline=[] with-0=[0]
card       body      LEAK    baseline=[] with-0=[0]
card       children  clean   (control fired: with-42=[42])
card       footer    LEAK    baseline=[] with-0=[0]
dialog     footer    LEAK    baseline=[Close] with-0=[0Close]
sheet      footer    LEAK    baseline=[Close] with-0=[0Close]
drawer     footer    LEAK    baseline=[] with-0=[0]
table      footer    LEAK    baseline=[H] with-0=[H0]

card children is the one clean row, and the reason is worth stating because it is not a guard: that site reads (schema.children || schema.body), and 0 || undefined evaluates to undefined. The || chain converts the falsy number away by accident. Author body: 0 on the same renderer with children absent and the same chain evaluates to 0 — which is the card body row directly above it. So the clean row is a coincidence of operand order, not protection.

Why these are authorable, not hypothetical

All of these slots are declared SchemaNode or SchemaNode | SchemaNode[] in @object-ui/types, and the zod mirror's node union carries a z.number() arm (nodeUnionOptions in packages/types/src/zod/base.zod.ts lists BaseSchemaCore, z.string(), z.number(), z.boolean(), z.null(), z.undefined()). The published validator therefore accepts children: 0; nothing between an author and this leak refuses it.

The census, with its population and its control

Instrument: the TypeScript checker, not a grep — for each JSX guard {X and-and …} in the tree, ask getTypeAtLocation(X) whether the type admits number.

• Population: 816 non-test .tsx files under packages/ and apps/ (enumerated with git ls-files); 2396 JSX and-and guards in them; 2396 left operands across those guards.
• Reported: 66 operands whose type admits number.
• Of those, typed SchemaNode / SchemaNode | SchemaNode[]: 14 sites — objectui#9033's header-bar rightContent, plus complex/table.tsx footer, layout/card.tsx (four operands), layout/container.tsx / flex.tsx / grid.tsx / stack.tsx children, overlay/dialog.tsx and overlay/sheet.tsx footer, and plugin-detail's DetailView.tsx header and footer.
• Control (must fire): header-bar.tsx — the one site already known to leak. It fired. A zero from an instrument with no lit control is not a reading.
• Indeterminate, reported rather than counted clean: 53 operands typed any / unknown, which carry no numeric type flag and so are invisible to a flag-based census. overlay/drawer.tsx footer and layout/containers.tsx's page:card body / footer are in this bucket — their schema is any — and the runtime probe above shows drawer footer leaking, so the bucket is not empty of real defects.

Not the same defect as objectui#8908, for the same reason objectui#9033 was not

toRenderableSchema has mapped falsy primitives onto nothing since objectui#8908. It is not reached: and-and short-circuits, so the chain produces the raw 0 before any bridge is called. renderChildren's own if (!children) return null first leg is bypassed identically — several of the sites above call it on the right-hand side and leak anyway.

The shape of the fix is already ruled

objectui#8331 settled it on DataTableSchema.emptyAction and objectui#9033 ported it to header-bar: replace the and-and chain with a ternary yielding null, keeping the truthiness leg. objectui#7105 governs the declaration side — node slots relax the renderer, they do not narrow the declaration.

Two things this card should decide rather than assume, because they are what make it bigger than a repeat of objectui#9033:

1. Is a per-site ternary the right remedy at this count, or does it want one guard? Eleven sites repaired one at a time is the same edit eleven times, and a twelfth slot added next month gets it wrong again. A shared renderNodeSlot(value) helper — or teaching renderChildren to be the guard and dropping the and-and entirely, since its first leg already answers null for every falsy input — closes the class instead of the instances.
2. plugin-detail's two sites are in a different package and were not probed here. They are census hits, not measurements.

Why it was filed rather than folded into objectui#9033's pull request

That card's dispatch named this exact outcome as a stop condition: report the list, fix the one line the card is about, do not silently widen. Eleven more renderers and a class-level remedy question is not a reviewable diff on a one-line card.

Refs

• objectui#9033 — header-bar's rightContent, the one site now fixed; this census came out of implementing it.
• objectui#8331 — the identical trap on DataTableSchema.emptyAction, and the ruled shape.
• objectui#8908 — the bridge repair. Adjacent symptom, different mechanism, reaches none of these.
• objectui#7105 — node slots relax the renderer rather than narrow the declaration.