The adapter-only data.* diagnostic (#5687) covers the visibility gate only — the same constant predicate on disabled greys a control out in silence

Found while wiring the disabled / disabledOn fault diagnostic (#6445). Filed unassigned — it is a second diagnostic with its own maintainer ruling and its own copy, not part of that card's wiring.

The gap

reportAdapterOnlyDataPredicate (packages/react/src/utils/visibilityDiagnostic.ts, #5687, ruling 2026-08-22 option A) reports a predicate that evaluated perfectly against the wrong object: at the node tier data is the data-source adapter, not the row, so data.status == 'draft' is a CONSTANT for every row. It is called from evaluateVisibilityPredicate's dev leg only.

The disabled / disabledOn gate — now wired for faults by #6445 — is not covered by it. A node written

{ "type": "button", "disabled": "data.status == 'locked'" }

evaluates cleanly (no fault, so #6445's reporter correctly stays silent), resolves undefined == 'locked' to false… and on the other value of that constant (data.locked == null, !data.assignee, or an adapter that answers nothing) hands the gate a constant true that greys the control out on every row, in every build, with nothing on the console.

Why it was not folded into #6445

Two reasons, both about the message rather than the wiring:

Its ruling is scoped to visibility. The 2026-08-22 option A ruling names the visibility gate; extending a dev-only diagnostic to a second gate is a decision, not an implementation detail.
Its copy is visibility copy. formatAdapterOnlyDataMessage says "on this surface a constant false hides the node on every row while looking exactly like a gate that said no" — the wrong sentence for a gate whose constant disables rather than hides, and the same asymmetry The disabled / disabledOn node gate has NO fault diagnostic in either build — and its fail-soft polarity greys a control out rather than showing it #6445 had to write new copy for.

So it needs a copy decision of the same kind #6445 made (its GATE_KIND_COPY table is the obvious place, and the module now has the machinery), plus a ruling on whether the extension stays dev-only as #5687 is.

Note on the deprecation clock

#5687's own docblock says this diagnostic "dissolves when #5330's deprecation window closes" (data.* as a spelling for the row). If that window is close, the right answer may be to do nothing here — worth stating explicitly rather than leaving the asymmetry undocumented, which is why this is filed rather than fixed.

Related: #5687 (the leg), #5330 (the deprecation), #6445 (the disabled fault diagnostic that exposed the gap).

Triage (triage seat, hourly round 2026-08-26, session session_01DoScS7Ei3iWc39AgGN36wU): graded needs-user-decision + domain:ui, type Task. Rationale: the card asks to extend a diagnostic whose existing scope is a recorded maintainer ruling (2026-08-22 option A, #5687, named the visibility gate only) to a second gate — extending a ruled scope is a decision, and the card's own deprecation-clock note (#5687 dissolves when #5330's window closes) means "do nothing" is a live option, not negligence. Filing into the decision box rather than the queue.

os-decision-facets

实际业务拉动:症状真实且完全静默 —— 一个 adapter-only 常量谓词把控件在每一行、每个构建里灰掉,console 一个字都不说;但发生率未测,且这是 dev-only 诊断面,不影响运行行为。
项目长远合理性:SchemaRenderer's node visibility gate resolves a data.* predicate against the data-source ADAPTER, so a hoisted properties.visible written that way hides the block on every row, silently #5687 自述随 [Decision] Row-predicate spelling contract: does the canon narrow to record.*, or do the bare shorthand and data.* spellings stay? (the uncarried "#4075 step 3" question) #5330 的 data.* 废弃窗口关闭而消解 —— 给一个将死拼写加第二条诊断,是对旧世界的投资;窗口远则值得,近则浪费。裁决应先看窗口时间表。
防 AI 犯错:常量谓词「完美求值于错误对象」正是 AI 写元数据 app 批量犯错被掩盖的形状;把它响亮化方向正确,且 The disabled / disabledOn node gate has NO fault diagnostic in either build — and its fail-soft polarity greys a control out rather than showing it #6445 已备好 GATE_KIND_COPY 机制,扩展成本是一条表项 + 一段 copy。
创业阶段不扩散:dev-only 诊断扩一个 gate 面很小,但每一次扩展都是 [Decision] Row-predicate spelling contract: does the canon narrow to record.*, or do the bare shorthand and data.* spellings stay? (the uncarried "#4075 step 3" question) #5330 收尾时要拆走的新义务;不扩(B)是零成本选项。
选项与推荐:A — 扩到 disabled gate,新 copy 进 GATE_KIND_COPY,保持 dev-only(推荐,条件:[Decision] Row-predicate spelling contract: does the canon narrow to record.*, or do the bare shorthand and data.* spellings stay? (the uncarried "#4075 step 3" question) #5330 废弃窗口不在近期关闭);B — 不扩,记录不对称,等窗口关闭一并消解(若窗口近,推荐 B);C — 扩且转 always-on(⛔ 超出 SchemaRenderer's node visibility gate resolves a data.* predicate against the data-source ADAPTER, so a hoisted properties.visible written that way hides the block on every row, silently #5687 先例,不推荐)。
置信缺口:[Decision] Row-predicate spelling contract: does the canon narrow to record.*, or do the bare shorthand and data.* spellings stay? (the uncarried "#4075 step 3" question) #5330 废弃窗口的时间表本分析未查(它是 A/B 之间的决定性事实);disabled 上 adapter-only 常量谓词的真实发生率未测。

Generated by Claude Code

Maintainer ruling recorded — Option A: extend the constant-predicate diagnostic to the `disabled` gate, dev-only

Provenance: maintainer, 2026-08-27, PM chat decision-inbox batch 5 (adjudication session session_01DKWDdUJ2XNRESVVWUvcpnh), verbatim: 「同意」 — accepting recommendation A with its stated premise affirmed: the #5330 data.* deprecation window is not closing near-term.

Ruled: extend the adapter-only constant-predicate diagnostic (#5687, 2026-08-22 option A) to the disabled / disabledOn gate — dev-only, via #6445's GATE_KIND_COPY machinery, with gate-appropriate copy (the constant-true direction: "greys the control on every row while looking exactly like a gate that said no"). ⛔ C (always-on) stays excluded — outside the #5687 precedent.

Inheritance clause, binding: the extension inherits #5687's dissolution note verbatim — when #5330's deprecation window closes, both legs dissolve together; the new copy entry must carry the same pointer so the teardown finds it.

State: needs-user-decision → pm:queue.

Generated by Claude Code

Claim: domain:ui @ objectui PM seat, session 8ca04858-ea8e-5b85-9182-de59aa49e00c — branch claude/issue-6504-disabled-gate-constant-predicate.

Dispatching on the maintainer's 2026-08-27 ruling (「同意」), Option A, unchanged: extend the adapter-only constant-predicate diagnostic to the disabled / disabledOn gate, dev-only, through #6445's GATE_KIND_COPY machinery.

⛔ C (always-on) stays excluded — outside the #5687 precedent. The extension is dev-only exactly as #5687 is.

⭐ The binding inheritance clause is the part most likely to be dropped

Quoting the ruling:

the extension inherits #5687's dissolution note verbatim — when #5330's deprecation window closes, both legs dissolve together; the new copy entry must carry the same pointer so the teardown finds it.

That is not decoration. #5687's own docblock says the diagnostic dissolves when #5330's data.* deprecation window closes. A second leg added without the same pointer is a second thing the teardown will miss — and a diagnostic that outlives the spelling it exists to police is exactly the kind of residue this lane spends its time removing. ⇒ The pointer is a deliverable, not a comment.

The copy is a decision the ruling already made — get its direction right

The existing message is visibility copy: "on this surface a constant false hides the node on every row while looking exactly like a gate that said no."

⛔ Wrong sentence for this gate. The ruling names the direction: the constant-true case, which greys the control on every row while looking exactly like a gate that said no. That asymmetry is the whole reason this was not folded into #6445 — same machinery, different sentence.

⚠️ And note the defect's shape, because it decides what the diagnostic must catch: the predicate evaluates perfectly, against the wrong object. At the node tier data is the data-source adapter, not the row, so data.status == 'locked' is a constant for every row. There is no fault, so #6445's fault reporter correctly stays silent — which is precisely why this needs its own leg rather than a widened one.

The premise the ruling rests on

The ruling affirms it explicitly: #5330's data.* deprecation window is not closing near-term. That was the decisive fact between A and B — triage flagged it as its own confidence gap and the maintainer answered it.

⛔ If you measure that the window is closing soon, that is a stop-and-report, not a reason to proceed quietly or to switch to B on your own. The seat re-decides; you do not.

Surface

packages/react/src/utils/visibilityDiagnostic.ts and #6445's GATE_KIND_COPY table. packages/react is free — PR #6690 (the settled-schema hook) merged and nothing else holds it.

⚠️ One card is in flight nearby: #6527's B patch round touches packages/types' exports map plus app-shell and plugin-designer. No overlap expected; report it if your work reaches any of those.

Type Task. Changeset patch or minor per the presence gate's own verdict — this is a dev-only diagnostic, no runtime behaviour and no published type changes, so ⛔ not clause-②. If your measurement says otherwise, stop and report.

Generated by Claude Code

os-dev-report

{
  "issue": 6504,
  "status": "done",
  "branch": "claude/issue-6504-disabled-gate-constant-predicate",
  "pr": "https://github.com/objectstack-ai/objectui/pull/6702",
  "premise_still_valid": true,
  "summary": "Extended objectui#5687's adapter-only `data.*` constant-predicate diagnostic to the `disabled` / `disabledOn` gate, dev-only, per the 2026-08-27 maintainer ruling (Option A) on this issue. Added a new `enablement` entry to a per-gate copy table (`ADAPTER_ONLY_GATE_COPY`, following #6445's `GATE_KIND_COPY` indexing discipline) with new copy for the constant-`true` direction (greys the control on every row), wired via a new dev-only branch in `evaluateEnablementPredicate` (`SchemaRenderer.tsx`) that reuses the single `onFault`-based engine call rather than paying for a second evaluation. Verified the stop-condition premise: #5330's deprecation window is not closing near-term (its Phase 2 survey carrier, #5741, is still open and unassigned, blocked on a live-tenant data channel), so proceeded per the ruling. The dissolution pointer is carried both at #5687's original docblock location (appended clause naming both legs) and in a new docblock on the copy table that quotes the ruling's binding clause verbatim and names every symbol + both call sites a #5330-window teardown must delete together.",
  "tests": "`pnpm --filter '@object-ui/react^...' build` (dependency closure: types/core/data-objectstack/i18n all Done) before any judgment. `pnpm --filter '@object-ui/react' run type-check` (tsc --noEmit && tsc -p tsconfig.test.json) -> exit 0, clean. `pnpm exec vitest run packages/react/src/__tests__/SchemaRenderer.disabledGateFaultDiagnostic.test.tsx` -> 'Test Files 1 passed (1)' / 'Tests 32 passed (32)' (23 pre-existing + 1 rewritten-from-negative-to-positive + 9 new). `pnpm exec vitest run .../SchemaRenderer.nodeGateDataPredicate.test.tsx .../SchemaRenderer.concealmentGateFaultDiagnostic.test.tsx` (unchanged siblings) -> 'Test Files 2 passed (2)' / 'Tests 40 passed (40)'. `pnpm exec vitest run packages/react/` (root-level invocation per objectui#3378) -> 'Test Files 62 passed (62)' / 'Tests 926 passed (926)', re-run after ablation restore with identical result. `node scripts/check-control-bytes.mjs` -> exit 0, 'OK (scanned 5553 tracked text file(s); skipped 85 binary)'. `node scripts/check-changeset-presence.mjs` -> exit 0, '3 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s)'. `node scripts/check-changeset-no-major.mjs` -> exit 0. Lint narrowed to the 3 touched files (`eslint --no-inline-config --format json`, exit captured before any pipe): 3 files linted per json array length, 0 errors, 18 pre-existing no-explicit-any warnings on SchemaRenderer.tsx, none within the edited line range (805-864) -- confirmed by running eslint via --stdin on origin/main@9abc02ab2's own copy of the file, which also reports exactly 0 errors / 18 warnings, proving zero new warnings. Narrowing justified: eslint.config.js declares no `project`/`projectService` (no type-aware linting), so the diff cannot move judgment on any untouched file. ABLATION, both legs predicted before running, mutation and restoration proven via git hash-object plus git diff HEAD (never an editor's exit code), each leg wrapped in a trap-based restore. Leg 1 (must-produce): reverted evaluateEnablementPredicate to its pre-change shape (call-site grep count 1->0, blob hash 57c0beb3f...->9f8e8dcd0...). Predicted every new positive-report test turns red and every silence/negative control stays green; observed exactly that -- 8/8 red (acceptance criterion, constant-true direction, copy-is-not-reused, cross-gate content, disabledOn, both dedupe cases, the rewritten pin), 24/24 other cases green. Restored: git diff HEAD empty, hash back to 57c0beb3f.... Leg 2 (false-positive direction): broadened the SHARED (pre-existing, unmodified-by-this-PR) discriminator regex DATA_ROOT_PATH_RE to also match `record` as a root (marker count 0->1, hash 40e466e7d...->191c9de53...). Predicted the new FALSE-POSITIVE CONTROL test and nodeGateDataPredicate.test.tsx's 3a would flip red. OBSERVED DIFFERENTLY, reported as observed rather than forced: neither predicted test flipped (both are independently protected by the `source.indexOf('data.') === -1` fast-reject, since neither predicate string contains the literal substring 'data.'); instead two unrelated PRE-EXISTING pins in nodeGateDataPredicate.test.tsx (2d, 3d) flipped, both already-hardened false-positive guards this card did not write, correctly catching the hostile mutation. Conclusion drawn from the miss: this leg adds no new discriminator logic, only a second call into the existing shared function, so its false-positive resistance is inherited by construction rather than separately proven -- a fact about the suite's (and the shared module's) discriminating power, not a gap in this leg. Restored: git diff HEAD empty, hash back to 40e466e7d.... Leg 3 (visibility leg unchanged): evaluateVisibilityPredicate's call site is untouched (still defaults to 'visibility', no gate argument added); confirmed both by the unmodified sibling test files passing byte-identical to origin/main and by leg 1's mutation showing zero effect on any visibility-leg test. Final tree confirmed clean (`git diff HEAD` 0 lines) after both ablation legs, at commit 20c52cedf (branch head, pushed) -- the sha the PR body's gate table and this report both quote against.",
  "mcp_calls": 8,
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

ACCEPT — PR #6702, head 20c52cedf, flipped ready for review.

What was verified, and what the dev found that the card did not predict.

1. The stop-condition premise holds. The card's dissolution pointer is only useful if #5330's deprecation window is not about to close under it. Checked rather than assumed: #5330's Phase 2 survey carrier, #5741, is still open, unassigned, and blocked on a live-tenant data channel. The window is not closing near-term, so a pointer (rather than an inline teardown) is the right shape here.

2. The dissolution pointer is carried in two places, not one.

Appended to SchemaRenderer's node visibility gate resolves a data.* predicate against the data-source ADAPTER, so a hoisted properties.visible written that way hides the block on every row, silently #5687's original docblock, naming both legs.
A new docblock on ADAPTER_ONLY_GATE_COPY that quotes the ruling's binding clause verbatim and names every symbol plus both call sites that a [Decision] Row-predicate spelling contract: does the canon narrow to record.*, or do the bare shorthand and data.* spellings stay? (the uncarried "#4075 step 3" question) #5330 teardown must delete together.

A pointer in one place is a pointer someone deletes half of. Both entry points a teardown could arrive through now carry it.

3. Leg 1 (must-produce): 8/8 red as predicted, 24/24 of the others green. The control behaves exactly as the card specified.

4. Leg 2's prediction missed — and the analysis is the more valuable half of this report.

Predicted: broadening the shared DATA_ROOT_PATH_RE to also match record would flip the new false-positive control and nodeGateDataPredicate.test.tsx case 3a red.

Observed: neither flipped. Both are independently protected by the source.indexOf('data.') === -1 fast-reject — neither predicate string contains the literal data., so the broadened regex is never reached. What did flip were two unrelated pre-existing pins (2d, 3d) — already-hardened guards this card did not write.

The dev reported this rather than reshaping the mutation until it produced the predicted signature. Its conclusion, quoted:

this leg adds no new discriminator logic, only a second call into the existing shared function, so its false-positive resistance is inherited by construction rather than separately proven — a fact about the suite's (and the shared module's) discriminating power, not a gap in this leg.

That reading is accepted. A leg that adds a second call into an already-guarded shared function inherits that function's resistance; demanding a separately-proven discriminator for it would be demanding proof of something the leg does not introduce. The correct record is the one written: the property is inherited, and stated as inherited.

5. Leg 3: the visibility leg confirmed unchanged, two independent ways.

Standing note for whoever executes the #5330 teardown: the two docblocks in (2) are a matched pair. Deleting one and leaving the other is worse than deleting neither — the survivor then points at a leg that no longer exists.

Generated by Claude Code

Navigation Menu

Repository navigation

The adapter-only `data.*` diagnostic (#5687) covers the visibility gate only — the same constant predicate on `disabled` greys a control out in silence #6504

The gap

Why it was not folded into #6445

Note on the deprecation clock

Maintainer ruling recorded — Option A: extend the constant-predicate diagnostic to the `disabled` gate, dev-only

⭐ The binding inheritance clause is the part most likely to be dropped

The copy is a decision the ruling already made — get its direction right

The premise the ruling rests on

Surface

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

The adapter-only data.* diagnostic (#5687) covers the visibility gate only — the same constant predicate on disabled greys a control out in silence #6504

Description

The gap

Why it was not folded into #6445

Note on the deprecation clock

Activity

os-support-ai commented on Aug 26, 2026

os-zhuang commented on Aug 27, 2026

Maintainer ruling recorded — Option A: extend the constant-predicate diagnostic to the disabled gate, dev-only

os-sales commented on Aug 28, 2026

⭐ The binding inheritance clause is the part most likely to be dropped

The copy is a decision the ruling already made — get its direction right

The premise the ruling rests on

Surface

os-sales commented on Aug 28, 2026

os-sales commented on Aug 28, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

The adapter-only `data.*` diagnostic (#5687) covers the visibility gate only — the same constant predicate on `disabled` greys a control out in silence #6504

Maintainer ruling recorded — Option A: extend the constant-predicate diagnostic to the `disabled` gate, dev-only