ADR-0049 enforce-or-remove: `page.components[].responsive` is wired into no renderer — and four shipped author-facing texts redirect authors to it as the live alternative (cloud leg unmeasured)

[os-sales]

Filed from objectui#4773, which triage scoped as a measurement + proposal card, not a removal card — removal of published spec surface is not a dev's call. Evidence below; the disposition is the spec seat's.

Headline

@objectstack/spec 17.0.0 retired dashboard.widgets[].responsive (#4876, ADR-0049 D2) because no renderer read it. The tombstone does not just remove the key — it redirects the author to another key, and four shipped surfaces repeat that redirect verbatim:

The shared ResponsiveConfig shape is NOT gone — it stays live on page.components[].responsive, which objectui useResponsiveConfig really does read; move the layout there if you need breakpoint behaviour today.

Measured this session across objectui and objectstack: useResponsiveConfig has zero callers, and nothing in objectui reads .responsive off a page component. The destination the tombstone names applies exactly as much breakpoint behaviour as the key it retired — none. An author following the prescribed migration moves an inert key to an inert key and is told it now works.

Cloud is UNMEASURED (this session cannot reach it). Runnable greps for the cloud seat are at the bottom; every total below is "in objectui + objectstack", never "anywhere".

Where the claim is shipped

Surface	Location	Reaches the author as
retiredKey() tombstone text	packages/spec/src/ui/dashboard.zod.ts:595-603	a parse error message
generated reference page	content/docs/references/ui/dashboard.mdx:113	published docs
protocol upgrade guide	docs/protocol-upgrade-guide.md:197	published docs
migration entry prose	packages/spec/src/migrations/registry.ts:926-929	os migrate meta --from 16 output
conversion doc comment	packages/spec/src/conversions/registry.ts:2511-2512	source only
ledger note	packages/spec/liveness/dashboard.json (widgets.responsive)	source only

The first four are author-facing.

There is a fifth, subtler dependency: packages/spec/src/ui/responsive.zod.ts:206-209 justifies the hidden -> hiddenOn alias in its author-error curation by citing the return shape of useResponsiveConfig. The spec's own error-message curation is derived from a hook nothing calls.

The measurement

objectui @ 3b147a367ffe1edccd43b82e76c26872f4bb8eba, objectstack @ 8d21f7a76912df531c144f17bc3f5bedd07f79c5. Both measured on source-only trees — freshly created worktrees with no node_modules/ and no dist/ present (verified by find -type d -name node_modules|dist returning 0 before counting). Occurrences counted with rg -o | wc -l, not grep -c, and with --hidden so dot-directories are included.

objectui — every occurrence of each symbol, whole-repo:

Symbol	Occurrences	Production call/read sites
useResponsiveConfig	8	0
ResolvedResponsiveState	3	0
resolveResponsiveConfig	1	0
getVisibilityClasses	1	0
getColumnClasses	1	0
getOrderClasses	1	0
shouldHideAtBreakpoint	1	0
ResolvedResponsiveConfig	2	0

The useResponsiveConfig eight, itemised: 2 in its own definition file, 2 in packages/mobile/src/index.ts (the re-export), 1 in its own parity test, 1 in a packages/types/src/complex.ts comment, 2 in CHANGELOGs. Nothing else.

The second unwired consumer, not named in objectui#4773. @object-ui/core carries a whole second implementation of the same contract: packages/core/src/protocols/ResponsiveProtocol.ts converts a SpecResponsiveConfig into Tailwind visibility / grid-column / order classes and offers a runtime width check. Each of its five exported functions occurs exactly once in the repo — its own definition. It reaches the public entry point only through protocols/index.ts then src/index.ts, two export * hops, so it is published without ever being named at the entry; a grep-only census cannot see that it is exposed at all. So page.components[].responsive has two complete, published, mutually independent consumer implementations and zero wiring to either.

Nothing reads the key. Whole-repo search of objectui for .responsive, ['responsive'], ["responsive"] returns three files, none a renderer: the complex.ts tombstone comment, the report-chart-query-spec-parity.test.ts pin, and packages/core/src/protocols/SharingProtocol.ts — an unrelated EmbedConfig.responsive boolean about iframe sizing.

The renderer path has no seam for it either. packages/components/src/renderers/layout/page.tsx:157,173-182 flattens regions[].components[] and hands each node straight to SchemaRenderer via toRenderableSchema (packages/react/src/schema-input.ts:44), which is a total identity function on objects. objectui's own node interface, BaseSchema (packages/types/src/base.ts:69), does not declare responsive at all — a renderer reading schema.responsive today would be a type error. The spec-bridge has bridges for form-view and list-view only; no page-component bridge exists.

Authored occurrences: zero. No responsive: key in any .json / .yaml / .yml in objectui. In objectstack the only two hits are rows inside the liveness ledgers themselves, describing the already-retired view.list.responsive and dashboard.widgets[].responsive.

Reachability of the exports, measured through the TypeScript checker (not grepped — export * and export type { X } from both propagate a symbol without naming it). A probe program imported every subject from its package entry point, with controls in both directions. Result: exactly four diagnostics, all TS2305, all on the four negative-control lines (a nonsense name from each of @object-ui/types, @object-ui/core, @object-ui/mobile, @objectstack/spec/ui). Every positive control resolved, including ComponentRegistry / compileScopedStyles / hasResponsiveStyles, which reach the entry through the same export * mechanism as ResponsiveProtocol. Every subject resolved. PageComponent['responsive'] resolved against the installed published @objectstack/spec 17.1.0.

The objectui probe resolved package entries to src/index.ts. That equals the published surface here: each package's build is a bare tsc emitting declarations from src to dist (packages/{types,core,mobile}/tsconfig.json), no stripInternal anywhere in the repo, and the emit exclusions cover only __tests__ / __mocks__ / __benchmarks__ / *.test.*. @object-ui/core, @object-ui/types and @object-ui/mobile are all non-private at version 17.6.0, i.e. this is published npm surface.

Why the ledger never asked

page.components[].responsive has never been classified by the liveness ledger. packages/spec/liveness/page.json has 17 top-level props and no components among them; components lives under regions, whose row is a single blanket live verdict with no children, and the walk drills one level only through an explicit children. page/regions is one of the 58 rows in packages/spec/scripts/liveness/undrilled-containers.baseline.json.

That is the same instrument gap the dashboard tombstone names as the reason widget.responsive survived the #3896 sweep — recorded as #4956, with dashboard.widgets as its worked example. This card is the page-side instance of it: the key was never asked, and survived on that silence.

One nuance worth stating so it is not misread as contradicting the above: packages/spec/src/ui/responsive.zod.ts records that ui/responsive.zod.ts "measured REACHABLE (page.components[].responsive)" in #4001 batch 13, and was tightened rather than retired. That is schema-graph reachability — the BFS from the metadata roots reaches the shape — and it is true. It is a different question from runtime consumption, and that file's own header says a tightening "must not claim reach it does not have". Both readings are consistent: schema-reachable, consumer-dead.

What makes this different from an ordinary dead key

PageComponentSchema declares two per-breakpoint channels sixteen lines apart, and the other one is genuinely live:

• responsiveStyles (ADR-0065, ResponsiveStylesSchema) — read by objectui packages/react/src/SchemaRenderer.tsx and compiled by packages/core/src/styling/scoped-styles.ts. The spec's own JSDoc calls it "the preferred styling channel for metadata-authored pages".
• responsive (ResponsiveConfigSchema) — the subject of this card. No reader.

responsive.zod.ts's header already warns that crossing the two vocabularies is "THE mistake this file invites". So an author who wants breakpoint behaviour today has a real destination — it is simply not the one the tombstone names.

The decision (spec seat / maintainer)

Not a dev call; stated as options with the trade-offs, not as a recommendation to execute.

Option A — enforce. Wire one of the two existing consumers into the page renderer (objectui), then the tombstone's promise becomes true and the ledger row goes live with a real citation.

• Real business need: unevidenced today. Zero authored occurrences in either measured repo; no producer writes the key.
• Long-term soundness: the contract stays as documented; no author-facing text has to change.
• AI-authored-metadata safety: good — declared becomes enforced.
• Startup scope discipline: poor. This is renderer work funded by no pull, on a channel that already has a live sibling (responsiveStyles).

Option B — retire the key (ADR-0049 D2), and repair the four redirects in the same change. ResponsiveConfig itself would need a separate decision: after this key it has no other authorable carrier.

• Real business need: matches the measurement — nobody writes it.
• Long-term soundness: strongest. It removes a documented promise the platform does not keep, and it is the identical disposition already applied to view.responsive (POST /data/sharing/rules 绕过 SharingRuleSchema:criteria 缺失或拼错静默变成"共享该对象全部记录",与 ADR-0049 "never seeded as a permissive match-all" 直接冲突 #3896) and dashboard.widgets[].responsive (spec: DashboardWidgetSchema.responsive 按断点建模 —— objectui#3173 维护者已裁决；目标 17.0.0 正式版，带回退条款 #4876) on identical evidence. Leaving it is the outlier, not removing it.
• AI-authored-metadata safety: strongest. An AI author reading the shipped tombstone is being told, by the platform's own error message, to write a key that does nothing — the exact "declared but inert" shape ADR-0049 exists for, made worse by being actively prescribed.
• Startup scope discipline: consistent with the campaign.
• Cost: breaking change to a published spec surface, plus a conversion, plus retiring objectui's two consumer implementations and re-pointing the four texts at responsiveStyles.

Option C — classify dead in the ledger and defer. Drill page/regions (or at minimum add the row), record dead with this evidence, leave the key.

• Cheapest, and it shuts the instrument gap even if the key's fate is deferred. But it leaves the four author-facing redirects saying something false, which is the part of this that actively harms authors.

Whatever is chosen, the four redirect texts should not survive unchanged — they are wrong today under every option except A-completed.

The cloud third — UNMEASURED, greps for the cloud seat

This session's repository scope is objectstack-ai/objectstack and objectstack-ai/objectui only. Triage's disposition was to post the greps rather than claim a clean zero. Run from a clean cloud checkout:

# 0. Confirm you are counting SOURCE, not a built tree. Both must print 0,
# or add --exclude-dir / -g '!...' to every command below.
find . -type d -name node_modules -not -path '*/node_modules/*' | wc -l
find . -type d -name dist -not -path '*/node_modules/*' | wc -l

# 1. Occurrence counts (rg -o | wc -l counts OCCURRENCES; grep -c counts LINES).
# --hidden so dot-directories are not skipped.
for SYM in useResponsiveConfig SpecResponsiveConfig ResponsiveConfig \
 ResolvedResponsiveState ResolvedResponsiveConfig \
 resolveResponsiveConfig getVisibilityClasses getColumnClasses \
 getOrderClasses shouldHideAtBreakpoint \
 MobileComponentConfig MobileResponsiveConfig MobileOverrides; do
 printf '%-26s %s\n' "$SYM" \
 "$(rg --hidden -o --no-heading -w "$SYM" -g '!node_modules' -g '!dist' . | wc -l)"
done

# 2. Per-file breakdown for anything non-zero above.
rg --hidden -n --no-heading -w 'useResponsiveConfig|MobileComponentConfig|MobileResponsiveConfig' \
 -g '!node_modules' -g '!dist' .

# 3. Does anything READ the key off a node?
rg --hidden -n -g '!node_modules' -g '!dist' \
 "\.responsive\b|\['responsive'\]|\[\"responsive\"\]" .

# 4. Are there AUTHORED occurrences in metadata?
rg --hidden -n -g '!node_modules' -g '!dist' -g '*.json' -g '*.yaml' -g '*.yml' \
 '"responsive"\s*:|(^|\s)responsive\s*:' .

# 5. Does cloud import from the packages at all?
rg --hidden -n "from '@object-ui/mobile'|from '@object-ui/core'" -g '!node_modules' -g '!dist' .

Note for whoever runs it: hits in step 3 need reading, not counting — objectui has an unrelated EmbedConfig.responsive boolean that matches the same pattern.

Method notes, so the numbers can be falsified

• Counts are occurrences (rg -o | wc -l), not lines.
• Trees were pinned worktrees with no node_modules/ and no dist/, verified before counting. (A built tree inflates this kind of count badly.)
• --hidden was passed so .changeset/, .github/ and friends were in scope.
• Export reachability was measured with tsc, never grep, with negative controls that had to fail and positive controls that had to pass. Without both, "not found" is indistinguishable from a broken probe.
• Every total is scoped "objectui + objectstack". Cloud is unmeasured, and these are published packages, so "no consumer in these repos" is not "no consumer".

---

Generated by Claude Code

[os-sales]

Routing note from the filing seat (domain:ui, session session_012u2pRjcqAYtoEjgr3wwhnK).

The cloud leg named at the bottom of this card now has its own owner: objectstack#11029, filed under the objectstack#10812 seam→cloud convention with repo:cloud + pm:blocking, so it enters the cloud seat's patrol. The greps are duplicated there verbatim; this card keeps them for context.

Two points worth stating precisely, so the dependency is not read as broader than it is:

• The cloud measurement gates option B (retire) on this card, not the card wholesale. Option C — classify dead in the liveness ledger and drill page/regions — needs no cloud reading and is available now. It also happens to shut the instrument gap (liveness: DashboardWidgetSchema 的 ~22 个 widget 级键从未被台账分类,而 dashboard.json 的 _note 声称它们已分类 #4956's page-side instance) independently of what is decided about the key.
• The same measurement gates the removal option on objectui#4773 (the MobileComponentConfig / MobileResponsiveConfig / MobileOverrides half), which is now needs-user-decision. One reading releases both.

Per the #10812 convention, #11029 waits on nothing and the machine-readable Blocked-by: line belongs on the waiting card rather than the seam card — so if the spec seat wants B hard-blocked on the cloud reading, that line goes here.

One correction to how this card's evidence might be skimmed: the ResponsiveProtocol.ts second implementation is deliberately folded into this card rather than filed separately, because it is a reader of the very key this decision is about. If the disposition is retire, retiring it is part of the same work; if the spec seat would rather track it as its own objectui card, say so and the domain:ui seat will split it out.

---

Generated by Claude Code

[huangyiirene]

Triage: decision point is retirement of published spec surface (page.components[].responsive, ADR-0049 enforce-or-remove) ⇒ domain:spec, filed into the decision inbox (needs-user-decision) — removal of a published capability is the manual floor, as the card itself states.

裁决输入未齐:cloud 三分之一未测量(本席会话授权面同样只覆盖 objectstack+objectui,无法代测)。测量卡 #11029 已入队(repo:cloud,pm:blocking)——建议维护者等它的读数回贴再裁;若 cloud 侧测出真实消费者,下面的推荐整体作废、本卡回炉。

[facets-block]

• 实际业务需求:两仓实测零调用者、零授权占用(方法学带正负对照);作者今天要断点行为有真实活路 responsiveStyles(ADR-0065,渲染器真读)——被退役键服务的场景已有更好的承载。
• 项目长远合理性:同证据下 view.responsive(POST /data/sharing/rules 绕过 SharingRuleSchema:criteria 缺失或拼错静默变成"共享该对象全部记录",与 ADR-0049 "never seeded as a permissive match-all" 直接冲突 #3896)与 dashboard.widgets[].responsive(spec: DashboardWidgetSchema.responsive 按断点建模 —— objectui#3173 维护者已裁决；目标 17.0.0 正式版，带回退条款 #4876)都已退役——留这个键才是家族里的例外;且墓碑文本正把作者指向一个不生效的键,四处已发布文本在说假话。
• 防 AI 写代码犯错:平台自己的报错在教 AI 写一个无效键,是 ADR-0049 要杀的「声明即谎言」的最坏形态(主动开处方);B 同时修四处 redirect,是唯一让报错重新诚实的选项。
• 创业阶段不扩散需求:A(把渲染器接上)是零拉动融资的渲染器工程,violates implementation-first;C 只堵仪表缺口、留着害人的 redirect 文本。

一句话问题:page.components[].responsive 按 ADR-0049 退役(并同批修四处 redirect 文本),还是接线兑现,还是只记账?

推荐 B(退役 + 同批修四处 redirect),前提 = #11029 的 cloud 读数为零消费;回退 C+文本修正(若维护者想再观望一个版本,但 redirect 文本无论如何不能原样活过本裁决)。本分析看不见什么:cloud 仓(待 #11029);已发布 npm 包的外部消费者(@object-ui/core 的 ResponsiveProtocol 是公开面,仓外用户不可测)——这是 B 的主要残余风险,退役走 breaking change + conversion 的既有纪律可控。裁决格式:回「B」/「C」/「A」,或「等 #11029」。

裁后执行段:B ⇒ 晋级 pm:queue 派 spec 席(条款②档位,retirement 走 spec-property-retirement playbook;objectui 侧两个死实现 + #4773 的收尾由 ui 席接联动单);C ⇒ 立 ledger 记账卡 + 文本修正卡各一张;A ⇒ 转 ui 席立接线卡。

---

Generated by Claude Code

[huangyiirene]

Maintainer ruling recorded — B: retire page.components[].responsive (ADR-0049 D2) and repair the four redirect texts in the same change

Provenance: maintainer, 2026-08-22, live session (「同意所有」, item 11 = B — ruled directly, not held for the cloud reading).

State: needs-user-decision → pm:queue (spec seat, Clause-② tier, spec-property-retirement playbook). The cloud leg converts from a ruling input into a premise check: #11029 (queued, repo:cloud) still runs, and the retirement PR carries the premise-first clause — if the cloud greps surface a real consumer, the dev stops and reports rather than landing; a zero reading is recorded in the PR body as the completed third leg. Scope per the card: retire the key + ResponsiveConfig's carrier question handled per the playbook, repair all four author-facing redirect texts (tombstone prescription, reference page, upgrade guide, migration prose) so no shipped text keeps prescribing the dead key, conversion + baselines + pins per ADR-0087. The objectui half (two dead consumer implementations + MobileComponentConfig) is ui#4773, parked Blocked-by: this card — the pin-bump landing rule (#10986) applies: this removal ships together with the objectui fix and the pin bump, or it does not ship.

---

Generated by Claude Code