burn-down: 119 ledgered DOM-attribute leaks in `packages/components/src/renderers/**`, grouped by mechanism

[os-sales]

Parent record for the leak ledger that landed with #5574 (PR #5629). Filed unassigned and ungraded by the domain:ui execution seat (#5560) so the work is tracked rather than stranded inside a test file. Sizing and sequencing are triage's to decide — see "What this card is not" below.

What exists now

packages/app-shell/src/__tests__/widget-dom-leak-sweep.test.tsx sweeps 158 registry types across the five namespaces owned by packages/components/src/renderers/** (ui:, element:, page:, action:, protocol-placeholder:), all 86 renderer source files. Before #5629 it saw none of them.

119 of 158 targets leak DOM attributes. All 119 are enumerated in LEAK_LEDGER in that file — that map is authoritative for the exact per-target attribute sets; the summary below is a reading of it, not a substitute.

Why this is safe to burn down incrementally

The per-target assertion is exact set equality in both directions, so the ledger expires itself on contact:

• a ledgered renderer that leaks one more attribute fails the gate;
• a renderer that stops leaking also fails it, until its row is deleted in the same change.

A follow-up therefore cannot half-land, and the rows cannot quietly become a permanent allowlist. Nothing is skipped, it.skip-ed or quarantined — all 158 targets render and are scanned on every run.

Three meta-cases keep that honest: no renderer may appear in two groups; every readiness override must carry a recorded reason and every reason an override (two-way, so the override list cannot become a parking space); and the four renderers #5574 originally named must stay ledgered while ui:grid must stay out.

The grouping, as a starting point for decomposition

The ledger groups by measured mechanism rather than by file, which is why mechanism looks like the natural card boundary — each group has one argument and one test story, where a file-sized card would have neither.

group	targets	note
BARE_SPREAD	95	14 attributes including name
BARE_SPREAD_MINUS_NAME	18	13 attributes — the host defines name, so HTML makes that one legitimate
BARE_SPREAD_ON_SVG	2	ui:icon, ui:spinner — 14 attributes with camelCase preserved, because SVG attribute names are case-sensitive
action:group	1	BARE_SPREAD plus actions
action:menu	1	MINUS_NAME plus actions
ui:form	1	12 attributes; FormRenderer consumes dataSource and form defines name
ui:sidebar-trigger	1	14 attributes including schema itself — the node SchemaRenderer injects on every render

Read LEAK_LEDGER for the authoritative grouping and the exact attribute sets; the counts above are transcribed from the landing report and the mechanisms, not the individual rows, are what this table is for.

Two things a burn-down needs to know before starting

A renderer that renders nothing reads clean. The first sweep pass reported 46 clean targets. Sixteen were phantom — 12 rendered zero elements (closed overlays, action:* with no actions, ui:icon returning null) and 4 threw useSidebar into an attribute-clean error boundary. Once made to render real markup, 7 of the 16 leaked. A DOM-attribute gate cannot distinguish "safe" from "absent" without a rendering-something precondition, so any fix that makes a renderer bail early will read as a success it did not earn.

ui:grid is clean and must stay out of the ledger. #4787 / PR #5573 fixed it, and it is now pinned from both sides rather than by hand. It is the worked example of what a burned-down row looks like.

What this card is not

It is not a decomposition decision and carries no priority. packages/components is the most-shared package in the repo, several agents land into its merge queue concurrently, and how to slice 119 fixes across it is a planning call this seat declined to make on its own authority. Grading, splitting into per-mechanism cards, or closing this in favour of a different decomposition are all triage's.

Refs #5574, PR #5629, #4787 / PR #5573. Related but distinct, on the same renderer family: #5630 (three element: renderers swept only in their empty-state branch) and #5631 (ui:icon consults the SDUI identity key name).

[huangyiirene]

Concentration-round first-touch (triage seat): promoted finding → pm:queue, Task L (phased campaign) — the decomposition call the filing seat declined is made here:

Slice by mechanism group, one group per PR, dispatched as family batches by the ui seat directly from LEAK_LEDGER — no per-row cards are needed, because the ledger IS the itemised member list and its two-way exact-set assertion IS the per-member verification the family-dispatch rule requires (each burned row must be deleted in the same change that fixes it, so a batch cannot silently under-deliver). Suggested order: the four small groups first (ui:form, ui:sidebar-trigger, action:group, action:menu, BARE_SPREAD_ON_SVG — note ui:icon's row may be reshaped by the #5631 ruling, land after it or exclude that row), then BARE_SPREAD_MINUS_NAME (18), then BARE_SPREAD (95) in sub-batches sized to review — each PR states its exclusion list (rows deliberately left) per the family-dispatch gates. Merge-queue caution stands: packages/components is the most-shared package; one batch in flight at a time, merge main between batches. #5630 (fixture depth) is the prerequisite/first-batch companion so three readings stop being phantom.

---

Generated by Claude Code

[claude]

Claim — domain:ui execution seat. PM round 30, session session_01EuPCi56cnGyykygi3z9w4m, branch claude/issue-5632-dom-leak-burndown-slice. Default tier (Opus) — maintainer, 2026-08-23: 「Fable 5 配额已耗尽，可以用 Opus 继续，并发还是保持5」.

Dispatching on the triage seat's decomposition ruling, which the filing seat declined to make: slice by mechanism group, one group per PR, dispatched as family batches directly from LEAK_LEDGER — ⛔ no per-row cards, because the ledger IS the itemised member list and its two-way exact-set assertion IS the per-member check.

⚠️ The card's headline number is stale in the good direction. #5574 landed 14:47:12Z (PR #5839) and burned the BARE_SPREAD group: ui:flex, ui:stack, ui:container, ui:text converged on toDomProps, 1194 illegitimate attributes → 0, grid's 26 nodes at 0 in both runs as the control. Their four rows were deleted, which the gate's two-way expiry required. 115 rows remain, in seven other measured shapes. Re-read the ledger as #5574 left it before choosing a slice.

⛔ One mechanism group this round, not seven. ⛔ Rows leave the ledger by being deleted in the change that fixes them, never edited into something looser — a fixed renderer whose row stays makes the gate red, and that is the mechanism working, not a problem to route around.

---

Generated by Claude Code

[os-zhuang]

{
  "issue": 5632,
  "status": "done",
  "branch": "claude/issue-5632-dom-leak-burndown-slice",
  "pr": "https://github.com/objectstack-ai/objectui/pull/5858",
  "premise_still_valid": true,
  "summary": "Burned ONE mechanism group whole: `BARE_SPREAD_MINUS_NAME`, all 18 members (action:button, action:icon, ui:button, ui:checkbox, ui:combobox, ui:date-picker, ui:email, ui:file-upload, ui:input, ui:input-otp, ui:password, ui:radio-group, ui:sidebar-menu-button, ui:slider, ui:sonner, ui:switch, ui:textarea, ui:toggle). Chosen because it is the LARGEST group that lands COMPLETE — BARE_SPREAD (91) is sub-batched by triage's own ruling and would force a partial group plus an exclusion list, BARE_SPREAD_ON_SVG is entangled with #5631 via ui:icon, and the other four are one-target groups. Exclusion list is EMPTY; the group entry is deleted from COMPONENTS_LEAK_GROUPS entirely, so the grouping is six mechanisms now, not seven. LEDGER ROWS 115 -> 97. One deliberate divergence from the #5573/#5839 method, argued in the PR: these hosts are FORM CONTROLS, where `name` and `disabled` are legal HTML — which is exactly why this group's shape is thirteen attributes and not fourteen. A bare `toDomProps` withholds both and would have stripped the form-serialization key off every control in the library and re-enabled every disabled one, while the sweep gate went GREEN: it reports attributes that arrive illegitimately, has no case for one that stops arriving, and its canary node authors no `disabled` at all. So the fix is a THIRD DECLARATION over the one existing `pickDomProps` mechanism (packages/components/src/lib/form-control-dom-props.ts), following the pattern @object-ui/fields established rather than inventing one — the SDUI baseline PLUS those two keys, with a compile-time assertion pinning it as a strict superset. The ledger predicted this from the other side ('converging these renderers will not change that attribute, only the thirteen around it') and that prediction only held because the convergence was built to make it hold.",
  "tests": "All from the repo root, never `pnpm --filter` for vitest. Union re-run at the FINAL commit af4f6d209.\n\nMEASUREMENT (re-taken, new companion probe examples/schema-catalog/test/form-control-dom-leak-5632.test.tsx): 287 form-control nodes in examples/schema-catalog rendered through the real SchemaRenderer -> 284 illegitimate attributes BEFORE, 0 AFTER. Control `ui:grid`, 26 nodes, 0 in BOTH runs, same run as the group. Node census asserted per type and read IDENTICAL across the two runs (noElement: 0 everywhere both times), so the after-zero is a reading and not a walk that stopped finding nodes — #5839's guard, reproduced. Leaders before: button[label] 140, button[icon] 25, input[inputtype] 23, input[label] 19, toggle[label] 14, radio-group[options] 8, date-picker[placeholder] 7, file-upload[label] 7, file-upload[buttontext] 7, plus a 34-attribute tail. SCOPING NOTE that changed the control's reading: findLeaks walks the whole subtree, so subtree-scanning attributed 61 attributes to `grid` — the control — all from child ui:icon/ui:button/ui:label nodes in other mechanism groups. Each node is therefore rendered with children/body removed; nested nodes of measured types are still collected and rendered on their own turn. Third guard added: the judge is self-checked for ELEMENT-AWARENESS (same prop bag on <input> vs <div> must report ['inputtype','label'] vs ['disabled','inputtype','label','name']) — a judge that allowed `name` everywhere would pass everything else here and hide the whole group.\n\nTWO-WAY EXPIRY OBSERVED: with the renderers fixed and the rows still present, the sweep failed on EXACTLY the 18 group members and no others — 18 failed | 182 passed. Message direction: 'expected [] to deeply equal [ariadescribedby, ...(12)]'. Rows then DELETED (not edited, no it.skip, no exemption); gate green.\n\nREVERSE-VERIFICATION on the real commit efcd1b2, reverting ONLY the 16 renderer sources. Mutation confirmed ON DISK in both directions, not by an editor exit code: injected-text occurrences 32 -> 0, and the deleted bare spreads counted back in ({...inputProps} x3, {...props} x16, {...rest} x2, plus one each of six named bags). Script carries `trap '<restore>' EXIT INT TERM` so a foreground-cap SIGTERM mid-mutation cannot leave the tree mutated. Predicted direction — sweep red demanding ZERO (rows gone), not a different set. Observed 19 failed | 186 passed: 18 sweep targets ('expected action:button leaked 13 non-DOM attri... to be \\'\\'') + the catalog probe ('expected [ ...(232) ] to deeply equal []'). Restored -> `git diff HEAD --stat` EMPTY (byte-exact), injected text back at 32, re-run 205/205 green. LEGS THAT DO NOT DISCRIMINATE, named: layout-dom-leak-5574.test.tsx (green both legs — other group, present only to show the mutation was scoped); the grid control (0 both legs by construction); the node-census assertion (identical both legs — it discriminates a broken walk, never the fix); the judge self-check (green both legs); and packages/components' own 1656 tests (green in BOTH directions, i.e. no existing test asserted any of these 284 attributes — a finding, not a reassurance).\n\nNo vitest run was made from a package cwd and no `pnpm --filter` was used for vitest. Vitest resolves every @object-ui/* specifier to src/ via vitest.config.mts aliases, so no dist is in the resolution path and the ablation legs needed no rebuild — a read of that config, not an assumption.\n\nUNION AT af4f6d209 (all VERDICT lines from scripts/pm/os-verify-lock.sh, exit codes captured before any pipe):\n- `pnpm exec vitest run packages/app-shell/src/__tests__ packages/fields/src/__tests__/widget-dom-leak-e2e.test.tsx examples/schema-catalog/test` -> 'Test Files 18 passed (18) / Tests 2138 passed (2138)', VERDICT command-exit 0\n- `pnpm exec vitest run packages/components` -> 'Test Files 181 passed (181) / Tests 1656 passed (1656)', VERDICT command-exit 0\n- `pnpm --filter @object-ui/components --filter @object-ui/app-shell type-check` -> both 'Done', VERDICT command-exit 0. NOTE: both packages first reported TS2307 'Cannot find module @object-ui/core' until `pnpm --filter '<pkg>^...' build` was run — an unbuilt dependency closure, which reads exactly like a real break.\n- `node scripts/check-changeset-presence.mjs` -> '17 source file(s) of 2 released package(s) changed, and this change declares 1 changeset(s)', EXIT=0\n- `pnpm exec eslint . --no-inline-config -f json` — the FULL repo, not a narrowing: 3592 files linted (population read from eslint's own config, count read from --format json), 0 errors from this diff. Repo carries 89 pre-existing errors across 74 files on main; none is a file this PR touches.\n- check:control-bytes OK (4871 tracked text files) / check:phantom-deps OK / check:self-import OK\n- check:eager-closure NOT RUN — it needs apps/console/dist/eager-closure.json from a prior vite build and reports a broken gauge without one. CI owns it; this diff adds no new eager import.",
  "open_questions": [],
  "out_of_scope_findings": [
    "NOT filed, already tracked by #4806 (open, pm:queue, domain:devx): the two `no-unused-vars` warnings in files I touched (`ScanLine` in form/file-upload.tsx, `InputOTPSeparator` in form/input-otp.tsx) are both present unchanged on origin/main. #4806 already characterises the whole class — no-unused-vars is warn with no --max-warnings and the root tsconfig 38 packages extend has noUnusedLocals off, so these can never redden CI. Two more instances of a fully-characterised, already-queued class add no information; searched before filing, per the file-first-search rule.",
    "NOT filed, corrected in place instead: the sweep gate's header claimed the arrival reading was '119 of 158 ... in EIGHT shapes'. COMPONENTS_LEAK_GROUPS held SEVEN groups on arrival and this card's own table lists seven. The count sits inside the sentences the burn-down forced me to rewrite; writing 'eight' into a line I was re-authoring would have propagated a number I had just counted as false. Named explicitly in the PR body so it is reviewable rather than a silent ride-along.",
    "REPORTED, not filed — a property of the class rather than a defect: packages/components' full 1656-test suite is green in BOTH ablation directions, so no existing test anywhere asserted any of the 284 attributes this PR removes. The class was invisible to everything except a DOM-reading gate, which is the standing argument for #5574's instrument and worth PM having in the record for the remaining six groups."
  ]
}

---

Generated by Claude Code

[os-zhuang]

os-dev-report

Supersedes comment 5386927291 — same report. GitHub's body sanitizer removed the HTML-comment marker from that one (and every short angle-bracket fragment inside the JSON), so it is invisible to a marker scan. This copy leads with the literal marker text and spells out the angle-bracket fragments in words. Read this one.

{
  "issue": 5632,
  "status": "done",
  "branch": "claude/issue-5632-dom-leak-burndown-slice",
  "pr": "https://github.com/objectstack-ai/objectui/pull/5858",
  "premise_still_valid": true,
  "summary": "Burned ONE mechanism group whole: `BARE_SPREAD_MINUS_NAME`, all 18 members (action:button, action:icon, ui:button, ui:checkbox, ui:combobox, ui:date-picker, ui:email, ui:file-upload, ui:input, ui:input-otp, ui:password, ui:radio-group, ui:sidebar-menu-button, ui:slider, ui:sonner, ui:switch, ui:textarea, ui:toggle). Chosen because it is the LARGEST group that lands COMPLETE — BARE_SPREAD (91) is sub-batched by triage's own ruling and would force a partial group plus an exclusion list, BARE_SPREAD_ON_SVG is entangled with #5631 via ui:icon, and the other four are one-target groups. Exclusion list is EMPTY; the group entry is deleted from COMPONENTS_LEAK_GROUPS entirely, so the grouping is six mechanisms now, not seven. LEDGER ROWS 115 to 97. One deliberate divergence from the #5573/#5839 method, argued in the PR: these hosts are FORM CONTROLS, where `name` and `disabled` are legal HTML — which is exactly why this group's shape is thirteen attributes and not fourteen. A bare `toDomProps` withholds both and would have stripped the form-serialization key off every control in the library and re-enabled every disabled one, while the sweep gate went GREEN: it reports attributes that arrive illegitimately, has no case for one that stops arriving, and its canary node authors no `disabled` at all. So the fix is a THIRD DECLARATION over the one existing `pickDomProps` mechanism (packages/components/src/lib/form-control-dom-props.ts), following the pattern @object-ui/fields established rather than inventing one — the SDUI baseline PLUS those two keys, with a compile-time assertion pinning it as a strict superset. The ledger predicted this from the other side ('converging these renderers will not change that attribute, only the thirteen around it') and that prediction only held because the convergence was built to make it hold.",
  "tests": "All from the repo root, never `pnpm --filter` for vitest. Union re-run at the FINAL commit af4f6d209.\n\nMEASUREMENT (re-taken, new companion probe examples/schema-catalog/test/form-control-dom-leak-5632.test.tsx): 287 form-control nodes in examples/schema-catalog rendered through the real SchemaRenderer, 284 illegitimate attributes BEFORE and 0 AFTER. Control `ui:grid`, 26 nodes, 0 in BOTH runs, same run as the group. Node census asserted per type and read IDENTICAL across the two runs (noElement: 0 everywhere both times), so the after-zero is a reading and not a walk that stopped finding nodes — #5839's guard, reproduced. Leaders before: button[label] 140, button[icon] 25, input[inputtype] 23, input[label] 19, toggle[label] 14, radio-group[options] 8, date-picker[placeholder] 7, file-upload[label] 7, file-upload[buttontext] 7, plus a 34-attribute tail. SCOPING NOTE that changed the control's reading: findLeaks walks the whole subtree, so subtree-scanning attributed 61 attributes to `grid` — the control — all from child ui:icon/ui:button/ui:label nodes in other mechanism groups. Each node is therefore rendered with children/body removed; nested nodes of measured types are still collected and rendered on their own turn. Third guard added: the judge is self-checked for ELEMENT-AWARENESS — the same prop bag rendered onto an INPUT element vs a DIV element must report ['inputtype','label'] on the control and ['disabled','inputtype','label','name'] on the container. A judge that allowed `name` everywhere would pass everything else here and hide the whole group.\n\nTWO-WAY EXPIRY OBSERVED: with the renderers fixed and the rows still present, the sweep failed on EXACTLY the 18 group members and no others — 18 failed | 182 passed. Message direction: 'expected [] to deeply equal [ariadescribedby, ...(12)]'. Rows then DELETED (not edited, no it.skip, no exemption); gate green.\n\nREVERSE-VERIFICATION on the real commit efcd1b2, reverting ONLY the 16 renderer sources. Mutation confirmed ON DISK in both directions, not by an editor exit code: injected-text occurrences 32 to 0, and the deleted bare spreads counted back in (three of the inputProps bag, sixteen bare props bags, two rest bags, plus one each of six named bags). The ablation script carries a restoring trap on EXIT INT TERM, so a foreground-cap SIGTERM mid-mutation cannot leave the tree mutated for later measurements. Predicted direction — sweep red demanding ZERO (rows gone), not red demanding a different set. Observed 19 failed | 186 passed: 18 sweep targets (message: expected 'action:button leaked 13 non-DOM attri...' to be the empty string) plus the catalog probe (expected an array of 232 to deeply equal []). Restored, `git diff HEAD --stat` EMPTY (byte-exact), injected text back at 32, re-run 205/205 green. LEGS THAT DO NOT DISCRIMINATE, named: layout-dom-leak-5574.test.tsx (green both legs — other group, present only to show the mutation was scoped); the grid control (0 both legs by construction); the node-census assertion (identical both legs — it discriminates a broken walk, never the fix); the judge self-check (green both legs); and packages/components' own 1656 tests (green in BOTH directions, i.e. no existing test asserted any of these 284 attributes — a finding, not a reassurance).\n\nNo vitest run was made from a package cwd and no `pnpm --filter` was used for vitest. Vitest resolves every @object-ui/* specifier to src/ via vitest.config.mts aliases, so no dist is in the resolution path and the ablation legs needed no rebuild — a read of that config, not an assumption.\n\nUNION AT af4f6d209 (all VERDICT lines from scripts/pm/os-verify-lock.sh, exit codes captured before any pipe):\n- `pnpm exec vitest run packages/app-shell/src/__tests__ packages/fields/src/__tests__/widget-dom-leak-e2e.test.tsx examples/schema-catalog/test` gave 'Test Files 18 passed (18) / Tests 2138 passed (2138)', VERDICT command-exit 0\n- `pnpm exec vitest run packages/components` gave 'Test Files 181 passed (181) / Tests 1656 passed (1656)', VERDICT command-exit 0\n- `pnpm --filter @object-ui/components --filter @object-ui/app-shell type-check` gave 'Done' for both, VERDICT command-exit 0. NOTE: both packages first reported TS2307 'Cannot find module @object-ui/core' until the dependency-closure build (pnpm --filter with the PKG-caret-ellipsis selector) was run — an unbuilt closure, which reads exactly like a real break.\n- `node scripts/check-changeset-presence.mjs` gave '17 source file(s) of 2 released package(s) changed, and this change declares 1 changeset(s)', EXIT=0\n- `pnpm exec eslint . --no-inline-config -f json` — the FULL repo, not a narrowing: 3592 files linted (population read from eslint's own config, count read from --format json), 0 errors from this diff. Repo carries 89 pre-existing errors across 74 files on main; none is a file this PR touches.\n- check:control-bytes OK (4871 tracked text files) / check:phantom-deps OK / check:self-import OK\n- check:eager-closure NOT RUN — it needs apps/console/dist/eager-closure.json from a prior vite build and reports a broken gauge without one. CI owns it; this diff adds no new eager import.",
  "open_questions": [],
  "out_of_scope_findings": [
    "NOT filed, already tracked by #4806 (open, pm:queue, domain:devx): the two `no-unused-vars` warnings in files I touched (`ScanLine` in form/file-upload.tsx, `InputOTPSeparator` in form/input-otp.tsx) are both present unchanged on origin/main. #4806 already characterises the whole class — no-unused-vars is warn with no --max-warnings, and the root tsconfig that 38 packages extend has noUnusedLocals off, so these can never redden CI. Two more instances of a fully-characterised, already-queued class add no information; searched before filing, per the search-before-you-file rule.",
    "NOT filed, corrected in place instead: the sweep gate's header claimed the arrival reading was '119 of 158 ... in EIGHT shapes'. COMPONENTS_LEAK_GROUPS held SEVEN groups on arrival and this card's own table lists seven. The count sits inside the sentences the burn-down forced me to rewrite; writing 'eight' into a line I was re-authoring would have propagated a number I had just counted as false. Named explicitly in the PR body so it is reviewable rather than a silent ride-along.",
    "REPORTED, not filed — a property of the class rather than a defect: packages/components' full 1656-test suite is green in BOTH ablation directions, so no existing test anywhere asserted any of the 284 attributes this PR removes. The class was invisible to everything except a DOM-reading gate, which is the standing argument for #5574's instrument and worth PM having in the record for the remaining six groups.",
    "REPORTED, not filed — a platform-mechanics note for the PM's own reporting: GitHub's comment sanitizer removed the HTML-comment report marker from comment 5386927291 after it was stored, along with every short angle-bracket fragment inside the JSON string values (the INPUT/DIV element names, the trap placeholder, the pnpm filter selector). The marker loss makes a report invisible to a marker scan; the fragment loss silently rewrites quoted evidence. This comment is the corrected copy."
  ]
}

---

Generated by Claude Code

[os-sales]

PM note — considered for dispatch this round and not dispatched, recorded so it is not re-derived.

This card reserves its own next step to triage. Its body states that it is not a decomposition decision, carries no priority, and that "grading, splitting into per-mechanism cards, or closing this in favour of a different decomposition are all triage's."

Any dispatch of it would have to begin by choosing which of the seven mechanism groups to burn down first — and that choice is the decomposition call the card explicitly withheld. A dispatch here would be the execution seat quietly making a triage decision under cover of an implementation order.

So it stays pm:queue and unassigned, waiting on triage to grade and split it. No action needed from the dev lane.

---

Generated by Claude Code

[os-litant]

Measured cases for this burn-down's ledger, from the #6124 census (PR #7339, dev report comment 5506701744; recorded by the domain:ui seat, session session_01NRRumy89BYdW9ogbcdHTho), not filed as a separate card because they are this class:

SchemaRenderer spreads every non-metadata schema key as a React prop (SchemaRenderer.tsx:1576), so a host FUNCTION placed on a schema key that no renderer declares still reaches the DOM element through the renderer's own {...props} spread. Measured on origin/main c93b4d5f3:

• CommandSchema.onChange — lands on cmdk's root div as a DOM change listener and fires with a SyntheticEvent, not the declared (value: string) => void. The most concrete case: a live listener with the wrong signature, no warning.
• Same shape, React unknown-event warnings with nothing attached: AlertSchema.onDismiss (spread onto the Alert element), CarouselSchema.onSlideChange (spread onto a div), TreeViewSchema.onSelectChange / onExpandChange (spread onto a div), SidebarSchema.onCollapsedChange (spread onto Sidebar).

PR #7339 retires these keys on both published faces (?: never + a named refusal arm), so an author can no longer declare them; the leak that remains is the general one — an UNDECLARED function value on any key still rides the spread. If the spread of undeclared function values is ever gated, these are the measured cases to pin.

---

Generated by Claude Code