`PALETTE_EXCLUSIONS` reasons say "no renderer" for two block types that do have registered renderers

[yinlianghui]

Observation filed unassigned while implementing #5837 (canvas chrome for renderable-but-unoffered block types). Not fixed there — #5837's change is keyed on renderer identity and deliberately leaves these two alone, so correcting the ledger text is a separate decision.

What was measured

packages/app-shell/src/views/metadata-admin/previews/block-types.ts groups four palette exclusions under the comment // No renderer, by decision. and gives each a reason string beginning "no renderer":

'ai:chat_window': 'no inline renderer — the floating chat overlay (plugin-chatbot) is canonical',
'element:form': 'no renderer — use the object-bound `object-form` block',
'element:record_picker': 'no renderer — record picking is a field widget, not a page block',
'element:text_input': 'no renderer — bare inputs belong to a form, not a page block',

Two of those four do have real, registered renderers today:

• packages/components/src/renderers/basic/text-input.tsx:161 — ComponentRegistry.register('text_input', ElementTextInputRenderer, { namespace: 'element', skipFallback: true, ... }), with a declared inputs list (objectui#3808) and a spec-declared defaultValue.
• packages/components/src/renderers/basic/record-picker.tsx:303 — ComponentRegistry.register('record_picker', ElementRecordPickerRenderer, { namespace: 'element', ... }), and that file's own comment already states the real reason: "element:record_picker is not in PUBLIC_BLOCKS (record picking is a field widget…)".

So the substantive half of each reason ("bare inputs belong to a form, not a page block" / "record picking is a field widget, not a page block") is still correct and still the decision. It is only the leading "no renderer" clause that is false, and core/src/registry/public-blocks.ts already words the same two exclusions without it.

ai:chat_window is genuinely unregistered — placeholders.tsx keeps it out on purpose so a referencing schema fails loudly — so its reason is accurate. element:form was not checked closely enough to claim either way here.

Why it is worth recording rather than leaving

block-config.test.ts enforces that every exclusion carries a reason (reason.length > 10); nothing enforces that the reason is true. That is exactly the drift shape PALETTE_EXCLUSIONS was made an explicit ledger to prevent (#2943): the ledger is read by later readers as the decision record, and #5837 had to re-derive registration state from source because the ledger's stated reason could not be trusted.

There is also a decision hiding behind the correction: if these two render, does the page canvas owe them real chrome the way it now owes the record:chatter alias one? #5837 says no — it keys on renderer identity within an alias group, not on bare registration, precisely so a block that is deliberately not page content keeps the generic box. Rewording the ledger should not be read as reopening that.

Shape of a fix, if wanted

Reword the two reason strings so the first clause states the actual decision (not page content / field widget) rather than a renderer status that is not true, and move them out from under the // No renderer, by decision. heading. Optionally check element:form the same way. Docs-only; no behaviour moves.

Refs: #5837 (where it surfaced) · #2943 (the guard that made PALETTE_EXCLUSIONS an explicit ledger).

[claude]

Concentrated triage batch: finding → pm:queue + domain:ui, Task, S — ledger-text truthfulness: two of the four PALETTE_EXCLUSIONS reason strings claim "no renderer" for types that HAVE registered renderers (measured). The exclusions themselves stand (they are decisions); correct the two reason strings to the true rationale (canonical-overlay / form-widget preference), so the ledger stops teaching a false fact to the next reader. Prose-only.

---

Generated by Claude Code

[yinlianghui]

Claimed + dispatched — domain:ui @ objectui execution seat

PM session session_01CSoz9uGhaaSgiq3hshtN7L, branch claude/issue-6071-palette-exclusion-reasons.
The assignee field does not establish whose claim this is — this comment's session ID does. No earlier claim comment exists.

裁决 (binding) — prose only

The exclusions themselves STAND. They are decisions, and ⛔ this card does not revisit them. Two of the four PALETTE_EXCLUSIONS reason strings claim "no renderer" for types that do have registered renderers. Correct those two reason strings to the true rationale (canonical-overlay / form-widget preference), so the ledger stops teaching a false fact to the next reader.

⛔ Do not remove an exclusion. ⛔ Do not add one. ⛔ Do not change any behaviour. If you find yourself editing anything but reason strings, you have left the card.

Verify

Re-derive which two of the four are false on today's main — confirm the renderer registration for each, and equally confirm the other two really are correct so you do not "fix" a true one. State all four verdicts in the PR body, not just the two you change.

⚠️ Get the replacement rationale from the code, not from triage's shorthand. "Canonical-overlay / form-widget preference" is triage's summary of why those types are excluded; the PR body should say what is actually true of each, in its own terms.

Traps

⚠️ This card has no behavioural pin available, and that is the finding. Reason strings are prose in a ledger — nothing parses them, so every gate is green before and after. Their green means you broke nothing; it is not evidence the new text is true. The verification is the renderer-registration measurement, and the PR body must say so.

⚠️ If a pin is feasible — e.g. asserting that no PALETTE_EXCLUSIONS entry whose reason says "no renderer" has a registered renderer — that would be worth more than the prose fix, because it stops the class recurring. Consider it and say why you did or did not add it. ⛔ Do not add one that merely restates the exclusion list.

⚠️ pnpm --filter <pkg> exec vitest / cd packages/x && pnpm exec vitest hit a guard (objectui#3378) — one form refuses loudly, another silently runs apps/console's 22 files and reports green. Always run vitest from the repo root; sanity-check the file/test count.

Standing rules

Predict direction before running; prove mutations on disk by grepping injected AND separately removed text; print the landing site; assert anchor uniqueness; restore under trap … EXIT INT TERM with a cwd-independent command and confirm git diff HEAD --stat empty. Derive gates by enumerating each CI job's own step list. Changeset: ⛔ let check-changeset-presence decide and quote its verdict line; ⛔ never major.

Collision bounds

Eight sibling agents are live. ⛔ Do not edit packages/plugin-form/**, packages/plugin-list/**, packages/plugin-detail/**, packages/types/**, packages/core/**, packages/plugin-kanban/**, packages/plugin-map/**, packages/app-shell/**, examples/schema-catalog/**, apps/site/**. ⚠️ #5935 is live and WIDE across packages/components — if PALETTE_EXCLUSIONS lives there, different files is fine, same file is not; check first and ⛔ report any file-level overlap rather than resolving it.

⛔ Leave the PR draft — the PM lands it.

---

Generated by Claude Code

[yinlianghui]

{
 "issue": 6071,
 "status": "blocked",
 "branch": "claude/issue-6071-palette-exclusion-reasons (created off origin/main ef2a3bd8d for measurement only; no commits, worktree and branch removed)",
 "pr": null,
 "premise_still_valid": true,
 "summary": "BLOCKED on a dispatch contradiction only the PM can resolve — no code written. PALETTE_EXCLUSIONS is defined in packages/app-shell/src/views/metadata-admin/previews/block-types.ts:149, and the dispatch collision bounds list a flat block on packages/app-shell/**. That path is the card's ONLY possible edit site (the prose fix), and it is also the only possible site for the optional pin (the precedent test palette-discussion-alias.test.tsx sits in the same previews/__tests__ directory). I did not resolve the conflict myself. MEASURED FILE-LEVEL OVERLAP IS ZERO: of the live app-shell siblings, #5935 touches previews/ActionPreview.tsx (same directory, different file), #6034 RecordDetailView.tsx + useConsoleActionRuntime.tsx, #6165 console/home/HomeRail.tsx, #5232 ObjectView.tsx; scanning every unmerged origin/claude/* branch for previews/block-types.ts, previews/block-config.ts or previews/__tests__/ returned exactly one hit, claude/issue-5837-canvas-icon-for-unoffered-types, and that is PR #6072 which was SQUASH-merged 2026-08-24T16:05:10Z (its content is already in main; the ref merely is not an ancestor). So the PM's one-line decision is: lift the app-shell bound for block-types.ts + a new file in previews/__tests__/, or re-dispatch when the app-shell siblings land. VERIFICATION IS COMPLETE AND DELIVERED HERE — all four verdicts re-derived on origin/main ef2a3bd8d. (1) ai:chat_window — reason TRUE: no namespace:'ai' registration exists anywhere under packages/*/src, placeholders.tsx:97 omits it deliberately so a referencing schema fails loudly. (2) element:form — reason TRUE in BOTH clauses (the issue left this one open): no namespace:'element' registration of 'form' exists; the only form renderers are ui:form (components/renderers/form/form.tsx:962, namespace 'ui') and view:form (plugin-form/src/index.tsx:184, namespace 'view', skipFallback), and the alternative its reason names IS real — plugin-form/src/index.tsx:125 registers 'object-form'. (3) element:record_picker — reason FALSE: registered at packages/components/src/renderers/basic/record-picker.tsx:303 with namespace 'element', skipFallback true. (4) element:text_input — reason FALSE: registered at packages/components/src/renderers/basic/text-input.tsx:161 with namespace 'element', skipFallback true. Decisive registry semantics: register(type, c, {namespace: n}) writes map key `n:type` (core/src/registry/Registry.ts:188) and get(type) with no namespace argument looks up that literal key (Registry.ts:375-384) — so element:text_input and element:record_picker resolve, element:form and ai:chat_window do not. Exhaustive enumeration on main: the only namespace:'element' registrations are definition-list, repeater, text, divider, image, button, number, metadata_viewer, record_picker, text_input. PROPOSED PATCH (2 strings, from the code — core/src/registry/public-blocks.ts:78-81 already words these same two exclusions without any renderer claim): element:record_picker -> 'renders, but not as page content — record picking is a field widget, not a page block (also excluded from PUBLIC_BLOCKS)'; element:text_input -> 'renders, but not as page content — bare inputs belong to a form, not a page block'; and move those two out from under the `// No renderer, by decision.` heading so it covers only ai:chat_window and element:form, which earn it. PIN VERDICT — feasible, and it changes the card's character, so I recommend it be authorized together with the prose fix: a test that reads each PALETTE_EXCLUSIONS reason and, for every entry whose text claims 'no renderer', asserts ComponentRegistry.get(type) is falsy. That is not a restatement of the exclusion list — it cross-checks reason PROSE against measured registry state and fails exactly on the drift this card is fixing. It is red on today's main for the two false entries, so it converts a prose-only card into a real red-to-green card, which is the evidence the dispatch says is unavailable. Feasibility is proven by precedent, not assumed: previews/__tests__/palette-discussion-alias.test.tsx already side-effect-imports a renderer package and asserts against the real ComponentRegistry from that exact directory. Two hazards the pin must carry: (a) an empty registry makes every negative assertion pass trivially, so it needs its positive half (after importing @object-ui/components, assert a known-registered sibling such as element:text is truthy) — the same discipline palette-discussion-alias.test.tsx states in its own header; (b) its coverage is bounded by which packages it imports, so a renderer landing in an unimported package would read as unregistered — the assertion should be scoped to that stated import set.",
 "tests": "No code changed, so nothing to gate; what I ran is the MEASUREMENT, and the dispatch's trap is confirmed rather than worked around. Run from the repo ROOT of a clean worktree at origin/main ef2a3bd8d (never via pnpm --filter): `pnpm exec vitest run packages/components/src/__tests__/palette-placeholder-blocks.test.tsx packages/components/src/__tests__/text-input-description-association.test.tsx packages/components/src/renderers/basic/__tests__/record-picker-label-association.test.tsx packages/app-shell/src/views/metadata-admin/previews/__tests__/block-config.test.ts` -> `Test Files 4 passed (4) / Tests 54 passed (54)`, EXIT=0. Exit code captured by redirecting to a file FIRST and reading tail afterwards, never through a pipe. File-count sanity check per the dispatch trap: 4 files requested, 4 ran — not apps/console's 22. Direction predicted before running: all green. What the green means: the two element tests throw 'element:record_picker is not registered' / 'element:text_input is not registered' if the lookup misses, so their passing is a RUNTIME confirmation that both types resolve from the registry; palette-placeholder-blocks.test.tsx:62 asserts ComponentRegistry.get('ai:chat_window') is falsy, a runtime confirmation of the opposite for that type. block-config.test.ts (the ledger's own suite) is green too — and that is the finding, not reassurance: it enforces only reason.length > 10, so it is green before and after any reason-string edit. No gate in this repo parses these strings, so gate colour is NOT evidence the new text is true; the registration measurement above is the evidence. No reverse verification was possible or attempted (no behavioural change to ablate) and none was faked. No install/build side effects left behind: node_modules removed, worktree removed without --force, empty local branch deleted.",
 "open_questions": [
 {
 "question": "The collision bounds forbid packages/app-shell/**, but PALETTE_EXCLUSIONS lives only there — so the card cannot be executed as bounded. Which way?",
 "options": [
 "A — Lift the bound narrowly: authorize packages/app-shell/src/views/metadata-admin/previews/block-types.ts plus one new file under previews/__tests__/, nothing else in app-shell. Measured file-level overlap with every live sibling and every unmerged branch is zero.",
 "B — Re-dispatch this card after the live app-shell siblings (#5935 previews/ActionPreview.tsx, #6034, #6165, #5232) land, keeping the bound as written.",
 "C — Prose fix only, no pin, under a narrow lift of block-types.ts alone."
 ],
 "recommendation": "A — file-level overlap is measured at zero (the only branch touching block-types.ts is #6072, already squash-merged), the bound as written appears to be a package-level listing that did not account for this card's own target file, and A is the only option that lets the pin land with the prose fix. The pin is what stops the class recurring, and it is red on main today, so it turns this into a red-to-green card instead of an unverifiable prose edit. B is safe but costs a round for a collision that does not exist at file level. C leaves the ledger correct today and unguarded tomorrow, which is precisely how the current false text got in."
 }
 ],
 "out_of_scope_findings": []
}

---

Generated by Claude Code

[yinlianghui]

os-dev-report

(Marker repost: in the previous comment the GitHub body sanitizer ate the <!-- os-dev-report --> HTML-comment marker after the write landed — read-back confirmed the stored body starts at the fenced block — so this copy leads with the literal text instead. Same report, verbatim; scan this one.)

{
 "issue": 6071,
 "status": "blocked",
 "branch": "claude/issue-6071-palette-exclusion-reasons (created off origin/main ef2a3bd8d for measurement only; no commits, worktree and branch removed)",
 "pr": null,
 "premise_still_valid": true,
 "summary": "BLOCKED on a dispatch contradiction only the PM can resolve — no code written. PALETTE_EXCLUSIONS is defined in packages/app-shell/src/views/metadata-admin/previews/block-types.ts:149, and the dispatch collision bounds list a flat block on packages/app-shell/**. That path is the card's ONLY possible edit site (the prose fix), and it is also the only possible site for the optional pin (the precedent test palette-discussion-alias.test.tsx sits in the same previews/__tests__ directory). I did not resolve the conflict myself. MEASURED FILE-LEVEL OVERLAP IS ZERO: of the live app-shell siblings, #5935 touches previews/ActionPreview.tsx (same directory, different file), #6034 RecordDetailView.tsx + useConsoleActionRuntime.tsx, #6165 console/home/HomeRail.tsx, #5232 ObjectView.tsx; scanning every unmerged origin/claude/* branch for previews/block-types.ts, previews/block-config.ts or previews/__tests__/ returned exactly one hit, claude/issue-5837-canvas-icon-for-unoffered-types, and that is PR #6072 which was SQUASH-merged 2026-08-24T16:05:10Z (its content is already in main; the ref merely is not an ancestor). So the PM's one-line decision is: lift the app-shell bound for block-types.ts + a new file in previews/__tests__/, or re-dispatch when the app-shell siblings land. VERIFICATION IS COMPLETE AND DELIVERED HERE — all four verdicts re-derived on origin/main ef2a3bd8d. (1) ai:chat_window — reason TRUE: no namespace:'ai' registration exists anywhere under packages/*/src, placeholders.tsx:97 omits it deliberately so a referencing schema fails loudly. (2) element:form — reason TRUE in BOTH clauses (the issue left this one open): no namespace:'element' registration of 'form' exists; the only form renderers are ui:form (components/renderers/form/form.tsx:962, namespace 'ui') and view:form (plugin-form/src/index.tsx:184, namespace 'view', skipFallback), and the alternative its reason names IS real — plugin-form/src/index.tsx:125 registers 'object-form'. (3) element:record_picker — reason FALSE: registered at packages/components/src/renderers/basic/record-picker.tsx:303 with namespace 'element', skipFallback true. (4) element:text_input — reason FALSE: registered at packages/components/src/renderers/basic/text-input.tsx:161 with namespace 'element', skipFallback true. Decisive registry semantics: register(type, c, {namespace: n}) writes map key `n:type` (core/src/registry/Registry.ts:188) and get(type) with no namespace argument looks up that literal key (Registry.ts:375-384) — so element:text_input and element:record_picker resolve, element:form and ai:chat_window do not. Exhaustive enumeration on main: the only namespace:'element' registrations are definition-list, repeater, text, divider, image, button, number, metadata_viewer, record_picker, text_input. PROPOSED PATCH (2 strings, from the code — core/src/registry/public-blocks.ts:78-81 already words these same two exclusions without any renderer claim): element:record_picker -> 'renders, but not as page content — record picking is a field widget, not a page block (also excluded from PUBLIC_BLOCKS)'; element:text_input -> 'renders, but not as page content — bare inputs belong to a form, not a page block'; and move those two out from under the `// No renderer, by decision.` heading so it covers only ai:chat_window and element:form, which earn it. PIN VERDICT — feasible, and it changes the card's character, so I recommend it be authorized together with the prose fix: a test that reads each PALETTE_EXCLUSIONS reason and, for every entry whose text claims 'no renderer', asserts ComponentRegistry.get(type) is falsy. That is not a restatement of the exclusion list — it cross-checks reason PROSE against measured registry state and fails exactly on the drift this card is fixing. It is red on today's main for the two false entries, so it converts a prose-only card into a real red-to-green card, which is the evidence the dispatch says is unavailable. Feasibility is proven by precedent, not assumed: previews/__tests__/palette-discussion-alias.test.tsx already side-effect-imports a renderer package and asserts against the real ComponentRegistry from that exact directory. Two hazards the pin must carry: (a) an empty registry makes every negative assertion pass trivially, so it needs its positive half (after importing @object-ui/components, assert a known-registered sibling such as element:text is truthy) — the same discipline palette-discussion-alias.test.tsx states in its own header; (b) its coverage is bounded by which packages it imports, so a renderer landing in an unimported package would read as unregistered — the assertion should be scoped to that stated import set.",
 "tests": "No code changed, so nothing to gate; what I ran is the MEASUREMENT, and the dispatch's trap is confirmed rather than worked around. Run from the repo ROOT of a clean worktree at origin/main ef2a3bd8d (never via pnpm --filter): `pnpm exec vitest run packages/components/src/__tests__/palette-placeholder-blocks.test.tsx packages/components/src/__tests__/text-input-description-association.test.tsx packages/components/src/renderers/basic/__tests__/record-picker-label-association.test.tsx packages/app-shell/src/views/metadata-admin/previews/__tests__/block-config.test.ts` -> `Test Files 4 passed (4) / Tests 54 passed (54)`, EXIT=0. Exit code captured by redirecting to a file FIRST and reading tail afterwards, never through a pipe. File-count sanity check per the dispatch trap: 4 files requested, 4 ran — not apps/console's 22. Direction predicted before running: all green. What the green means: the two element tests throw 'element:record_picker is not registered' / 'element:text_input is not registered' if the lookup misses, so their passing is a RUNTIME confirmation that both types resolve from the registry; palette-placeholder-blocks.test.tsx:62 asserts ComponentRegistry.get('ai:chat_window') is falsy, a runtime confirmation of the opposite for that type. block-config.test.ts (the ledger's own suite) is green too — and that is the finding, not reassurance: it enforces only reason.length > 10, so it is green before and after any reason-string edit. No gate in this repo parses these strings, so gate colour is NOT evidence the new text is true; the registration measurement above is the evidence. No reverse verification was possible or attempted (no behavioural change to ablate) and none was faked. No install/build side effects left behind: node_modules removed, worktree removed without --force, empty local branch deleted.",
 "open_questions": [
 {
 "question": "The collision bounds forbid packages/app-shell/**, but PALETTE_EXCLUSIONS lives only there — so the card cannot be executed as bounded. Which way?",
 "options": [
 "A — Lift the bound narrowly: authorize packages/app-shell/src/views/metadata-admin/previews/block-types.ts plus one new file under previews/__tests__/, nothing else in app-shell. Measured file-level overlap with every live sibling and every unmerged branch is zero.",
 "B — Re-dispatch this card after the live app-shell siblings (#5935 previews/ActionPreview.tsx, #6034, #6165, #5232) land, keeping the bound as written.",
 "C — Prose fix only, no pin, under a narrow lift of block-types.ts alone."
 ],
 "recommendation": "A — file-level overlap is measured at zero (the only branch touching block-types.ts is #6072, already squash-merged), the bound as written appears to be a package-level listing that did not account for this card's own target file, and A is the only option that lets the pin land with the prose fix. The pin is what stops the class recurring, and it is red on main today, so it turns this into a red-to-green card instead of an unverifiable prose edit. B is safe but costs a round for a collision that does not exist at file level. C leaves the ledger correct today and unguarded tomorrow, which is precisely how the current false text got in."
 }
 ],
 "out_of_scope_findings": []
}

---

Generated by Claude Code