feat(types)!: retire six dark ChatbotSchema keys as ADR-0049 tombstones (objectui#7703) - #8154

✅ Console Performance Budget

The eager closure is every chunk the entry reaches through static imports — what the browser fetches and parses before the app renders. The entry chunk on its own is a small fraction of it.

📦 Bundle Size Report

Size Limits

• ✅ Core packages should be < 50KB gzipped
• ✅ Component packages should be < 100KB gzipped
• ⚠️ Plugin packages should be < 150KB gzipped

Contract review — PASS — PR #8154 (objectui#7703)

Reviewed-by: domain:ui @ objectui PM seat, session_01YBWFb5YgMU5dw8p2VKj16S, in-seat at opus = TIER_DEFAULT. CONTRACT_REVIEW_TIER (claude-fable-5-1) is measured unavailable (429); the exemption permits falling back to default and ⛔ no lower. ⛔ Not described as running at contract-review tier.

⚠️ First: this PR repairs a miss in MY review of #8077. Recording that before anything else.

The dev declared two files beyond the surface my claim listed. I verified the justification and it is not merely acceptable — it is cleaning up after me.

content/docs/plugins/plugin-chatbot.mdx carries three PRESENT-TENSE claims that are false on today's tree. Verified by this seat:

:156  "unfiltered props spread, so some keys its type does not declare"
:192  "an authored value currently reaches that panel only through the registration's unfiltered props spread (objectui#7708)"
:193  (the same shape, for processVisibility)

But the fence is in place — toDomProps(props) at renderer.tsx:110, :298 and :449, so all three registrations now filter at the head of the element. And git show --stat d3499b315 (PR #8077, which I reviewed and landed) shows a changeset, two test files, renderer.tsx and complex.ts — no doc.

⇒ #8077 falsified three published-doc claims and I passed it without checking. objectui#7070's rule is 「restated, not left to rot」 and I did not enforce it. ⭐ The dev found it, declared it, and fixed it. ⛔ That is not scope creep; that is the rule being applied by the next seat because the previous one skipped it.

⚠️ One correction to the dev's own reading, immaterial to the conclusion: it describes d3499b315 as 「one changeset, two test files, one renderer」 — it is five files, also packages/types/src/complex.ts. The load-bearing half (⛔ no doc touched) is exactly right.

The second "beyond the list" file was not beyond it. packages/types/src/__tests__/chatbot-registration-authoring-faces-7655.test.ts:60 says verbatim 「showAvatars are LIVE on a chatbot-floating node today」 — false since #8077 — and that file is in packages/types, inside the dev's own fence. ⇒ it over-declared conservatively, which is the right direction to err.

⭐ And it correctly did not touch .changeset/7655-…md:43, which carries the same stale sentence: a landed changeset is historical record, ⛔ not a document to retrofit.

① Derived judgments

⭐ The census was rebuilt with the card's own instrument, ⛔ not carried from my dispatch. My Zone-2 number was a whole-src count and I flagged it as not per-registration. The dev split renderer.tsx at its three ComponentRegistry.register calls (bodies 60–240 / 241–378 / 379–531) and counted schema.KEY per body: all six keys 0/0/0.

Six lit controls, and three of them asymmetric — placeholder 1/1/1, messages 1/1/1, userAvatarUrl 1/1/1, maxHeight 1/1/0, floatingConfig 0/0/1, processVisibility 0/1/0. ⇒ the asymmetric three prove the instrument separates registrations rather than merely returning non-zero somewhere. A symmetric-only control set would have passed while the splitter was broken. That is the difference between a control and a reassurance.

Enforce-or-remove: all six REMOVE, and the argument is structural rather than per-key taste. The chatbot registration renders the plain Chatbot component (index.tsx:20-31), whose props are messages, placeholder, onSendMessage, disabled, showTimestamp, userAvatarUrl, userAvatarFallback, assistantAvatarUrl, assistantAvatarFallback, maxHeight — not one of the six. ⇒ every enforce arm would mean either growing a component prop (a feature, not a repair) or publishing a second spelling of a live key (AGENTS.md #0.1). Each key still gets its own migration target: userAvatar→userAvatarUrl, assistantAvatar→assistantAvatarUrl, markdown→enableMarkdown, height→maxHeight (or floatingConfig.panelHeight), loading→runtime state the chat runtime owns.

⭐⭐ The showAvatars argument is the sharpest thing in this PR. Declaring it on the two faces that do reach ChatbotEnhanced would re-open by declaration the channel objectui#7708 closed by fence. ⇒ the enforce arm for that one key is not merely unhelpful, it would undo a landed ruling through the type system. Easy to miss, and it is the reason this key needed a decision of its own rather than batching with the other five.

The provenance sentence is carried as adjudicated, in all three places I asked for — changeset, tombstone comment, and pin: showAvatars is 「a key the fence turned dark」, ⛔ not 「a key nothing ever read」. Backed rather than asserted: ChatbotEnhanced.tsx:489 really declares the prop and reads it at :1368 / :3097-3114 / :3295; the other five are not ChatbotEnhancedProps members at all, so the old spread had nothing to land them on.

No parity ledger row, with the reason: a ?: never member is still a declared key and a retirementTombstone() arm is still a mirrored key, so the pair stays in lockstep (the #7129 / #7779 route). Green in the suite with ⛔ no edit to the parity file — which is the evidence, not the claim.

② semver

@object-ui/types: minor with the break spelled out. ✅ major forbidden by convention; Changeset Bump Policy green.

③ Boundary flags

• Ablation, both legs, red-first. Leg 1 (zod tombstone removed): exactly 4 of 40 tests fail, all showAvatars ⇒ the pin is per key, not a blanket. Leg 2 (TS face): TS2578: Unused '@ts-expect-error' directive at :344 / :377 ⇒ the directives are real enforcement, not decoration. Restore proven by blob-hash equality against the recorded HEAD blobs plus an empty git diff HEAD — ⛔ not by an exit code. Restore pointed at HEAD explicitly rather than a bare git checkout -- (which would take the polluted index) — a real trap, correctly avoided.
• ⭐ The stale-dist hazard was addressed by argument, not skipped: the pin imports both faces relatively and the root vitest config aliases @object-ui/types to src, so no build sits between mutation and measurement. Stated rather than assumed — which is exactly the phrase that separates a checked hazard from an unexamined one.
• ⭐ Lint was the FULL repo scan (Tasks: 47 successful, 47 total, 0 errors) ⇒ no narrowing declared at all. ⭐ And turbo run type-check whole-repo 81 successful — the downstream-consumer sweep a narrowing owes, since retiring keys can only break consumers. The dev also verified empirically that the prefix filter ...@object-ui/types selects 43 dependents while the suffix form selects 2. ⇒ it checked that its own filter meant what it thought; a wrong filter here would have produced a green that measured almost nothing.
• Two gates hit preconditions first and were reported as NOT MEASURED, then resolved and green — check:doc-snippets (PRECONDITION NOT MET, exit 2, resolved by building the closure the gate names) and check:readme-exports (exit 1 = unbuilt-package on the untouched plugin-ai). ✅ Neither was reported as a pass while unmeasured, and neither was reported as a red.
• ⛔ processVisibility correctly NOT folded in — and better, it is retained as one of the asymmetric live controls (0/1/0) and still validated. Scope held: registry.ts untouched (the finding(types): SchemaRegistry / ComponentType omit chatbot-enhanced and chatbot-floating — two registered keys the layer can now name #7704 region fence), BaseSchema untouched (finding(types): BaseSchema's [key: string]: any leaves every component schema open, so a "declare the surface" fix can never reject a misspelled TOP-LEVEL key #5155).
• open_questions: none. ⭐ The report also notes that a label read-back 「is not decisive」 and that a delayed re-check is owed — the correct epistemic status for a write it cannot observe settle.

Verdict

PASS. needs:contract-review is attached (confirmed on three read-back legs) and is cleared in this stroke, with this comment as its provenance.

⛔ Not landing yet. CI unconverged: 32 enumerated against total_count: 32, 8 still in_progress (Lint, Test shards 1–4, Type Check, Build Docs, Live E2E), zero failures. ⚠️ Lint and Build Docs are both unread, and Build Docs matters more than usual here because this diff edits an .mdx. Live E2E will fail; its signature will be checked against objectstack#16186's before it is stood down, ⛔ not assumed.

Generated by Claude Code

Landing — PR #8154 (objectui#7703), head 3ca9d4477, 2026-09-06T21:2xZ.

Pre-landing three:

① Contract review PASS (comment 5562189869) at opus; needs:contract-review cleared with that as provenance. Head unchanged since the review.

② CI converged, enumerated 32 against total_count: 32 — 28 success · 3 skipped · 1 failure, ⛔ zero in_progress.

Read explicitly, because each mattered for a different reason on this diff:

• ⭐ Lint green (21:09:02Z) — the standing rule after R1's miss.
• ⭐ Build Docs green (21:08:13Z) — this diff edits content/docs/plugins/plugin-chatbot.mdx, so this is the job that could have caught a broken page. Doc Snippet Type Check, Doc Fence Language Check and Internal Docs Link Check are green too.
• ⭐ Test (shard 4/4) green (21:15:48Z) — worth naming: that is the shard carrying scripts/__tests__/, and it is where the sibling PR fix(data-objectstack): discriminate a refused mapping read from a served zero, on a channel #8152 was caught adding a fourth site for a known TranslateFn collision. This diff adds no exported name, and the gate agrees.

The single failure is Live E2E (informational) (job 101558421062).

⚠️ Stated precisely, because it is a narrower read than the previous three: I matched the downstream half of the signature — the no such table: sys_organization / sys_user reads, the one-per-minute WARN [recipients] email 'admin@objectos.ai' lookup failed … keeping verbatim retry loop, ##[error]Process completed with exit code 1, and the informational-lane summary. I did not re-read the createLocalAccountIssuer line at the head of the dump on this job specifically.

⇒ that is still a sufficient discriminator, and here is why: the window contains no Playwright output and no test failure — the job died at backend readiness, before any objectui code was exercised. A failure this PR caused would have to appear after that point and would look nothing like this. Carded objectstack#16186; the red is by design per objectui#7689, whose triage ⛔ forbids reverting the pin. ⛔ No re-run spent.

③ Ready flipped, auto-merge SQUASH enabled (⚠️ the API echoes an empty method even when it applied — verify by queue ref and parent-count). #8153 is ahead in the queue; this stacks behind it.

⭐ Landing control established BEFORE the queue takes it, on both faces:

⇒ both zeros are readings on a live file, and both faces are checked because a tombstone that landed on only one of them is exactly the half-state this card exists to prevent.

Fixes #7703, so the card closes on merge and this seat strips pm:dispatched and clears the assignee in the same stroke.

Generated by Claude Code