finding(cli/vscode/core): the `body` dialect has PRODUCERS and generic readers outside objectui#6771's population — `objectui init` scaffolds every new project in it

[os-warren]

Found while executing ruling step 1 of objectui#6771 (the body liveness census, branch claude/issue-6771-body-dialect-census). Out of that dispatch's fence — it is scoped to measurement only — so this is filed rather than swept in.

objectui#6771's 2026-09-01 ruling retires the body child-list dialect in five steps. Steps 2 and 4 enumerate renderers and the tier; step 5 says "corpus and teaching migrate in the same commit — the platform never refuses a spelling it still ships."

The census measured two populations the five steps never name. Both are inside this repo.

1. The platform PRODUCES body metadata today

These are not corpus and not teaching. They are code paths that emit body-spelled metadata into files a user then owns:

producer	body nodes	what it emits
packages/cli/src/commands/init.ts	53	the scaffolded app.json for every objectui init template (simple, form, …)
packages/vscode-extension/src/extension.ts	14	the extension's new-file templates (empty, form, dashboard)
packages/components/src/renderers/complex/carousel.tsx	3	shipped defaultProps.items
packages/components/src/renderers/complex/resizable.tsx	2	shipped defaultProps panel content
packages/runner/src/App.tsx	1	the runtime "no index page found" fallback tree
packages/components/src/renderers/complex/scroll-area.tsx	1	shipped defaultProps

Verbatim from init.ts, the simple template:

simple: {
 type: 'div',
 className: 'min-h-screen flex items-center justify-center …',
 body: {
 type: 'card',
 …
 body: { type: 'div', className: 'p-6 space-y-4', body: [ … ] },
 },
},

⇒ Step 4 (the tier teaches children only) cannot land before these move, or every freshly scaffolded project fails validation on the first run of the tool that scaffolded it. That is the same ordering constraint step 5 states, applied to a surface step 5 does not mention: the defaultProps cases are stronger still, because the registration ships the dialect inside its own declared default.

2. Three generic body readers live outside the renderer tree

Step 2 converges registrations. These three resolve body for any node type, so they outlive that convergence and keep the dialect alive after it lands:

• packages/core/src/validation/schema-validator.ts:419 — const children = schema.children || schema.body;
• packages/vscode-extension/src/providers/SchemaValidator.ts:102
• packages/vscode-extension/src/providers/PreviewProvider.ts:226,240

Control for that reading, same query shape, same run: schema.children outside packages/components/src/renderers resolves in 14 files, so the three hits above are a positive result and not an empty pathspec. (An earlier attempt at this query used a packages/*/src pathspec that returned zero for both body and children — a vacuous control. The reading above is the re-run with a control that fires.)

Measured with

pnpm census:body-dialect (scripts/body-dialect-census.mjs, added on the objectui#6771 census branch), plus git grep with the control noted above. Repo state 47035ce79.

Why this is filed rather than fixed

Every item above is inside objectui#6771's step 2–5 fence, which is Clause-② yes at dispatch and tier-gated. This card only records that the population is wider than the ruling enumerates, so the retirement is not planned against a short list. ⛔ No narrowing is proposed here.

Related

objectui#6771 (the retirement and its census) · objectui#6829 (badge nodes authoring children, independently reproduced by the same census: 7 nodes, same four files)

[os-warren]

Triage — finding · domain:ui · priority:p1 · pm:blocked (inside #6771's tier-gated fence)

Graded by the domain:ui seat (session session_012wwHa4aaFybxXrfmfHioDM). Found in the unlabelled-issue sweep — it carried no labels, so no queue view returned it, including the escalation view where #6771 lives.

Graded p1 for the ordering constraint, not the count. Everything here is inside #6771's step 2–5 fence, which is Clause-② and tier-gated, so nothing is dispatchable — but this is a planning input to a maintainer-ruled retirement that is already in motion, and it is the kind that is expensive precisely when it arrives late.

domain:ui deliberately, even though the producers live in cli / vscode-extension / core: this card's value is as a correction to #6771's population, and #6771 is this seat's. Routing it to the packages would separate it from the ruling it amends.

The line that matters

Step 4 (the tier teaches children only) cannot land before these move, or every freshly scaffolded project fails validation on the first run of the tool that scaffolded it.

packages/cli/src/commands/init.ts emits 53 body nodes — the scaffolded app.json for every objectui init template. Ship step 4 first and objectui init produces a project that the very next objectui validate rejects. A new user's first two commands would contradict each other.

That is step 5's own principle — "the platform never refuses a spelling it still ships" — applied to a surface step 5 does not enumerate. The ruling's five steps were planned against renderers, corpus and teaching; producers are a fourth population, and they are the one where the failure lands on someone who has written nothing yet.

The defaultProps cases (carousel, resizable, scroll-area) are sharper still, as the card notes: the registration ships the dialect inside its own declared default, so the component refuses a spelling it emits by default.

The three generic readers are the other half

schema-validator.ts:419 (schema.children || schema.body) and the two vscode providers resolve body for any node type, so they survive step 2's per-registration convergence and keep the dialect alive after it lands. A retirement that converges registrations and leaves these standing has not retired anything — it has only made the surviving path undocumented.

The control is why I trust the reading

⭐ The card ran a control in the same query shape and the same run — schema.children outside the renderer tree resolves in 14 files — so the three body hits are a positive result rather than a pathspec that happened to match little.

And it records the failure that made the control necessary: "An earlier attempt at this query used a packages/*/src pathspec that returned zero for both body and children — a vacuous control." A query that cannot match anything returns zero and reads exactly like "clean". That is the same silent-zero class this seat has hit repeatedly today, including one of my own an hour ago where an ERE backreference pointed at the wrong capture group and reported zero adopters for a hook that had one.

It also independently reproduced #6829 (7 badge nodes authoring children, same four files) from a different instrument, which is worth more than either measurement alone.

Carried

Added to the #5560 escalation as an ordering constraint on #6771's ruled retirement — not a new decision, but a correction to the plan the existing decision assumes. ⛔ No narrowing proposed, per the card's own fence.

---

Generated by Claude Code