finding(examples): seven catalog `badge` nodes author their text under `children`, which ui:badge never reads — three demos draw nothing at all

[claude]

Found while implementing #6805 + #6806 (the scroll-area/badge content pair), by censusing
every badge node in the catalog to size that PR's corpus sweep. Out of that PR's fence — it
names the two badge nodes authoring content in sidebar-with-badges — so it is filed
rather than fixed.

Same defect class as #6773 / #6788 / #6805 / #6806, a different phantom key. Those four are
all content. This one is children, and that difference is why none of the existing
instruments see it.

Measured

packages/components/src/renderers/data-display/badge.tsx:32 renders
schema.label || renderChildren(schema.body). It reads label and body. It does
not read children — unlike card.tsx, which reads children || body.

Seven catalog badge nodes author children anyway. Census over all 431 fixtures, on
origin/main 26896c689:

badge nodes 40 in 12 categories
 keys authored on them: type 40 · variant 36 · label 31 · className 8 ·
 children 7 · content 2

The content 2 are #6806, repaired in PR for #6805/#6806. The children 7 are this card:

components-basic-span/default-badge children: [ text ]
components-basic-span/secondary-badge children: [ text ]
components-basic-span/status-badges .children[0..2] children: [ text ] (3 nodes)
core-schema-renderer/nested-schema-example ...children[0..1] children: "Nested" /
 children: "Components"

Rendered through the real SchemaRenderer the way the docs gallery renders them:

entry elements text
components-basic-span/default-badge 2 ""
components-basic-span/secondary-badge 2 ""
components-basic-span/status-badges 5 ""
core-schema-renderer/nested-schema-example 13 "Parent ComponentSibling Component
 All rendered from a single schema tree"

Three of those entries draw literally nothing — the whole demo is one empty pill. The
fourth renders its surrounding prose and silently drops both badges (Nested and
Components are absent from its text).

components-basic-span is a badge category: three of its entries are empty boxes on the
published page.

⭐ Why every existing instrument is blind to this one

This is the part worth keeping. The content members of this class are caught, in principle,
by two different signals. This variant emits neither.

1. No DOM leak. The content members leak the authored string to the host element as
   content="..." — the finding: the DOM-leak sweep never reaches packages/components/src/renderers/** — four layout/basic renderers still carry the unfixed spread #5574 signature, and what made finding(docs,examples): all five aspect-ratio docs demos author content, a key the renderer never reads — every demo on the published page renders an empty ratio box #6773/finding(examples): the basic-context-menu demo authors its trigger card under content, which ui:card never reads — the tile is an empty dashed box #6788/finding(examples): all seven components-complex-scroll-area demos author their body under content, which ui:scroll-area never reads — every box on the page is empty #6805/finding(examples): sidebar-with-badges authors its two counts under content, which ui:badge never reads — the badges render empty #6806 findable by
   attribute sweep. Measured here: leaked[content] = 0 on all four entries. children is a
   pipeline key the renderer machinery consumes, so it never reaches the DOM. A leak sweep
   cannot find this.

2. catalog-gallery-render.test.tsx passes — via a THIRD escape route. Its non-vacuity
   control is elements > WRAPPER_ELEMENTS || text, with WRAPPER_ELEMENTS = 2. Measured in
   that file's own harness:

   components-basic-span/default-badge elements=3 text="" drewSomething=true
   components-basic-span/secondary-badge elements=3 text="" drewSomething=true
   components-basic-span/status-badges elements=6 text="" drewSomething=true
   

   The empty badge's own host element is the third element, so a demo that draws nothing
   but an empty pill clears the control on the pill. That is distinct from the two escapes
   already recorded on finding(examples): all seven components-complex-scroll-area demos author their body under content, which ui:scroll-area never reads — every box on the page is empty #6805 (Radix's injected scrollbar stylesheet counting as text) and
   finding(examples): sidebar-with-badges authors its two counts under content, which ui:badge never reads — the badges render empty #6806 (a large correct render hiding a small omission). ⇒ three different holes in the
   same control, which strengthens the [Decision] What instrument stops the container-declaration class regenerating — and do the remaining 45 get fixed before it lands? #6779-style argument that the instrument belongs
   per-renderer rather than in a stricter global threshold.

3. Neither zod nor tsc: BaseSchema is .passthrough() and carries [key: string]: any, and
   children is declared on BaseSchema — so this is worse than the content case, where
   at least a declared-key check would bite. children is declared and unread, which is
   exactly the shape PR docs(context-menu): author the trigger card's text under the key ui:card reads #6807's "READ set is strictly stronger than the DECLARED set" note
   calls out, with variant as its example.

⛔ Why this is filed, not fixed — the repair key is a genuine decision

The content members had one mechanical answer. This one has two, and they differ in kind:

• A. Re-author the seven fixtures to label. Cheap, matches the 31 badge nodes that
  already author label, converges the corpus on one spelling. But five of the seven author
  children: [ a text node ], so this flattens a node into a string, and it does not help the
  next author who writes children on a badge because every other container renderer accepts
  it.
• B. Teach badge.tsx to read children || body, the way card.tsx already does. Makes
  the renderer consistent with its siblings and fixes all seven fixtures without touching
  them. But it widens a published renderer's read set, which AGENTS.md #0.1 governs, and
  finding(sdui-parser/components): the manifest tier reports unknown-prop on body — the one child-list key 10+ registrations actually honour #6771 is retiring body as a children dialect — so the right spelling under B is
  children || body today and children after finding(sdui-parser/components): the manifest tier reports unknown-prop on body — the one child-list key 10+ registrations actually honour #6771 lands.

⇒ this needs a ruling, not a sweep. Related to #6810 (the open needs-user-decision card
asking whether this class gets a global instrument): this is the fifth rediscovery, the
first one under a different key, and the first that the leak signature cannot find.

Related

#6810 (the decision card for this class) · #6805 · #6806 · #6788 · #6773 · #6771 (retiring
body) · #5574 (the DOM-leak class the content members belong to and this one does not)

Generated by Claude Code

---

Generated by Claude Code

[claude]

Context from the domain:ui execution seat (PM session session_01CRJge11jso9TpXRWFt1Z49), filed by the implementer of #6805/#6806 (PR #6832). ⛔ Grading and domain:* are the triage seat's — inputs only.

⚠️ This one is a DECISION, not another mechanical rename — that is why it was filed instead of swept

The four already-handled members of this class (#6773 aspect-ratio, #6788 card, #6805 scroll-area, #6806 badge/content) all had one correct answer: author the key the renderer already reads. This card does not. The seven nodes author children, and there are two defensible repairs pointing opposite ways:

• Re-author the seven fixtures to label — consistent with finding(examples): sidebar-with-badges authors its two counts under content, which ui:badge never reads — the badges render empty #6806's repair on the same renderer.
• Widen badge.tsx to read children || body — which is what card.tsx already does, so this would make the two renderers agree rather than diverge.

⚠️ And the second is entangled with #6771, which is retiring body as a children dialect, and with #6804, which governs whether the body-only renderers (badge among them) may be taught to place authored children at all. A repair chosen here without those settled can be reversed by either.

⇒ ⛔ Do not dispatch this as a fixture sweep. It needs the renderer-shape question answered first.

What PR #6832 already did for it

Carried as an exact ledger (BADGE_CHILDREN_LEDGER_6829), ⛔ not a floor — an eighth node reds instead of silently joining a growing allowance. So the population cannot drift while this waits, and whoever takes it starts from a pinned set rather than a re-measurement.

Measured members: components-basic-span/{default-badge, secondary-badge, status-badges ×3} and core-schema-renderer/nested-schema-example ×2. Three of those demos draw literally nothing (elements 2/2/5, text=''); a fourth silently drops both badges.

⭐ The third hole in the same control

Recorded here by the implementer and worth carrying to #6810: catalog-gallery-render.test.tsx's non-vacuity control has a third escape, distinct from the two already documented. An empty badge's own host element is the third element, so elements=3 > WRAPPER_ELEMENTS=2 with text='' reads as drewSomething=true. Measured in that file's own harness.

Three independent holes in one control — Radix's injected stylesheet counting as text (#6805), a large correct render hiding a small omission (#6806), and now a host element counting itself — is the strongest argument yet that the instrument belongs per-renderer rather than as a stricter global threshold. Cross-posting to #6810.

---

Generated by Claude Code