finding(plugin-detail): RelatedList.unmaterializedSort.test.tsx:153 rides a SECOND, ungated settle — its gate covers a different promise chain, and it is green only by effect-declaration order

[os-sam]

Filed by the domain:ui execution seat out of #6967's plugin-detail slice. Measured there with a probe, carried in #6967's closing comment so it would not leave with that card, and filed here now that a dedupe was possible.

The defect

packages/plugin-detail/src/__tests__/RelatedList.unmaterializedSort.test.tsx:153 asserts:

expect(screen.queryByRole('button', { name: /Total/ })).toBeNull()

That assertion depends on objectSchema, which arrives from a different promise chain than the one the block's gate covers — getObjectSchema, via the effect at RelatedList.tsx:430. The block's gate settles the fetch chain. Nothing gates the schema chain.

The measurement

With getObjectSchema held pending:

• the block's gate passes
• Quantity is present
• Total is STILL present — because withheldFromServerSort falls through to isUnmaterializedFieldType(undefined), which is false

⇒ :153 goes RED.

It is green today only because that effect is declared before the fetch effect and both mocks resolve on the first microtask. Nothing pins that ordering, and nothing declares it as a requirement.

⚠️ This is a THIRD class — do not fold it into either existing one

• Not [measured out] async-settled surfaces read with a SYNCHRONOUS query — 21 sites triaged across 4 packages, ZERO defects; #6959 was the only one #6967's class (a synchronous DOM read of an async-settled surface). That card was measured out: 21 sites across 4 packages, zero defects, and its final predicate is recorded there as property (3).
• Not the absence/vacuous-green class ([measured out] async-settled surfaces read with a SYNCHRONOUS query — 21 sites triaged across 4 packages, ZERO defects; #6959 was the only one #6967's 57 subtracted sites), which fails by passing when it should not. This one fails by going red when nothing is wrong.

The distinguishing fact is that the block has a gate, and the gate is simply pointed at the wrong chain.

Why it is worth a card rather than a note

A flaky RED does not merely annoy — it ejects unrelated PRs from the merge queue. #6959 is the measured precedent: one racing test ejected two PRs that do not touch its package at all, one of which passed and failed on byte-identical content. That is the cost being carried here.

Suggested first step

⛔ Not a rewrite. Establish first whether the correct remedy is to gate the schema chain as well, or to make the assertion independent of it. #6959's own fix is the precedent worth reading before choosing: it kept the existing gate and added a second one, because the two gates prove different facts, and trading one for the other would have silently weakened the test.

Dedupe bound, stated rather than implied

Full-text issue search is unavailable to this seat (the endpoint has been refusing all day). The dedupe performed was a title-level scan over all 97 open pm:queue cards, plus the 8 open cards carrying finding. No match. ⚠️ That bound does not cover cards outside pm:queue, nor a card that mentions this site only in its body. If this duplicates something, close it against the original — the miss is the seat's.

Related: #6967 (the census this came out of, measured out), #6959 (the ejection precedent and the gate-substitution trap).

---

Seat session: https://claude.ai/code/session_013hfmP9hoMd3dJwTh85J4yB

[claude]

Claim — os-dev seat, session session_013hfmP9hoMd3dJwTh85J4yB.

Branch: claude/issue-7007-relatedlist-second-settle (pushed, empty, before the first edit).
Base: 592acafbeed97c5aed81a18f4d848b6179d3c044.

Declared file surface for this claim:

• packages/plugin-detail/src/__tests__/RelatedList.unmaterializedSort.test.tsx — the gated blocks
• .changeset/*.md — empty-frontmatter, test-only

No runtime source is expected to change. packages/plugin-detail/src/RelatedList.tsx will be mutated temporarily for the subject ablation and restored from HEAD; it is not part of the diff.

First step is measurement, not an edit: reproduce the red at :153 on the base spelling by mutating the timing fact only (delaying the getObjectSchema settle), then choose between gating the schema chain and making the assertion independent of it.

---

Generated by Claude Code

[claude]

os-dev-report

{
 "issue": 7007,
 "status": "done",
 "branch": "claude/issue-7007-relatedlist-second-settle",
 "pr": "https://github.com/objectstack-ai/objectui/pull/7076",
 "pr_url": "https://github.com/objectstack-ai/objectui/pull/7076",
 "premise_still_valid": true,
 "remedy_chosen": "Gate the schema chain as well. Each of the two affected blocks KEEPS its existing fetch-chain gate and ADDS a second gate (settleObjectSchema) that settles the getObjectSchema chain: it awaits the same promise the component awaited and flushes the commit through React.act. The assertions themselves are unchanged.",
 "why_not_the_other": "Making the assertion independent of the schema chain is not available: the value it reads IS derived from the fetched schema (withheldFromServerSort reads objectSchema.fields[field]; with nothing there it falls through to isUnmaterializedFieldType(undefined) === false). Removing that dependency would stop the block measuring #3950 at all. Establishing this was the first step, by measurement, before any edit.",
 "6959_precedent_read_and_followed_or_diverged": "READ (commit c0c436db8 + its changeset) and FOLLOWED, not diverged. #6959 kept mount()'s find-call gate and added an async read because the two gates prove different facts. Same here: gate 1 (view committed WITH ROWS) is what keeps the queryByRole(...).toBeNull() read non-vacuous - proven by ablation 4 below - and gate 2 proves the schema landed. Also followed its explicit trap: the added gate is NOT expect(getObjectSchema).toHaveBeenCalled(), since a mock CALL is issued one resolution before its value reaches state; toHaveBeenCalled is used only to obtain the promise.",
 "flake_reproduced_before_fix": {
 "how": "On the unmodified base spelling (HEAD 592acafbe), the TIMING FACT was mutated and nothing else: makeDataSource's getObjectSchema changed from `vi.fn(async () => objectSchema)` to a 50ms-delayed resolve. Mutation proved on disk before the run (original-line count 1 -> 0, marker count 0 -> 1); file restored from HEAD afterwards with a blob-hash comparison (fa8ca3306...) and an empty `git diff HEAD`.",
 "observed": "vitest exit 1. `Tests 2 failed | 3 passed (5)`. :123 `AssertionError: expected true to be false` (columnSortable('total')); :153 `AssertionError: expected [the Total button] to be null` with the full button DOM dumped. The blocks' own gates passed and the stored-column controls (Quantity / Unit Price / Sequence) were present in the same run - i.e. exactly the card's measurement. After the fix, the identical 50ms probe gives `Tests 5 passed (5)`; at 1500ms (beyond RTL's 1000ms waitFor budget) still `5 passed`, with file test time rising to 3.15s, which distinguishes a real settle from a wider polling window."
 },
 "test_not_weakened": {
 "ablation_method": "Two ablations, both against the FIXED test file, both mutating packages/plugin-detail/src/RelatedList.tsx (committed subject, restored from HEAD afterwards; each mutation proved on disk by anchor/marker grep counts plus `git diff --numstat` = `1 1`, each restore proved by blob-hash equality with HEAD and an empty `git diff HEAD`; scripts carried `trap restore EXIT INT TERM` with absolute paths). (1) SUBJECT ablation: withheldFromServerSort's live arm `return isUnmaterializedFieldType(fieldDef)` -> `return false`. (2) VACUITY ablation: the sort-button row's render condition forced false, to check the negative assertion cannot pass by absence. No rebuild was needed or possible to skip: the test imports ../RelatedList relatively inside the same package, so no dist sits in the resolution path - and both ablations turning the file red is itself proof the mutated source reached the runtime.",
 "still_red_on_subject_break": "YES. Subject ablation: `Tests 2 failed | 3 passed (5)` - :164 `expected true to be false` and :197 `expected [the Total button] to be null` (the card's :153 and its sibling :123, shifted by the added helper). Vacuity ablation: the block fails at its POSITIVE control instead - `TestingLibraryElementError: Unable to find an accessible element with the role \"button\" and name /Quantity/` at :193 - so the toBeNull() read cannot pass vacuously."
 },
 "checks_run": [
 {
 "name": "package tests",
 "command": "pnpm exec vitest run packages/plugin-detail/",
 "result": "PASS - `Test Files 118 passed (118)`, `Tests 1089 passed (1089)` (run under the shared verify lock; VERDICT line reported INNER_VITEST_EXIT=0)"
 },
 {
 "name": "type-check",
 "command": "pnpm --workspace-concurrency=2 --filter '@object-ui/plugin-detail^...' build, then pnpm run type-check in packages/plugin-detail",
 "result": "PASS - `TYPECHECK_OK`. First attempt failed with TS2307 'Cannot find module @object-ui/core' because the dependency closure was unbuilt, not because of this diff. `tsc -p tsconfig.test.json --listFiles` shows the edited test file IS in the program (1 hit), so this reading covers the edit."
 },
 {
 "name": "lint",
 "command": "pnpm run lint in packages/plugin-detail (plain eslint, no --no-inline-config)",
 "result": "PASS - `891 problems (0 errors, 891 warnings)`; all warnings pre-existing, the edited file contributes none"
 },
 {
 "name": "changeset presence",
 "command": "node scripts/check-changeset-presence.mjs",
 "result": "PASS - 1 changeset with EMPTY frontmatter, 'declared as releasing nothing, which is the explicit exemption'"
 },
 {
 "name": "changeset shape",
 "command": "pnpm changeset:check",
 "result": "PASS - 'No changeset declares a `major` bump.'"
 },
 {
 "name": "control bytes",
 "command": "pnpm check:control-bytes",
 "result": "PASS - 'scanned 5861 tracked text file(s)'; plus a manual grep -naP scan of the two changed files, no hits"
 },
 {
 "name": "vi.mock gates",
 "command": "pnpm check:vi-mock-specifiers && pnpm check:vi-mock-inherit",
 "result": "PASS - both OK"
 },
 {
 "name": "vitest invocation form",
 "command": "root-relative path filter from the repo root throughout",
 "result": "PASS - never `pnpm --filter <pkg> test`, never a path after `--`"
 }
 ],
 "claim_comment_url": "https://github.com/objectstack-ai/objectui/issues/7007#issuecomment-5481182785",
 "github_api_ok": true,
 "blockers": [],
 "summary": "Established by measurement that the assertion's value cannot be made independent of the getObjectSchema chain (it IS the subject), so the remedy is a SECOND gate, not a swap - following #6959. Both refusal blocks keep their fetch-chain gate and add settleObjectSchema(), which awaits the promise the component awaited and flushes through React.act. The card names :153; the sibling block at :123 was measured to carry the identical defect in the same file and is fixed in the same commit (declared in the PR body with its evidence). Test-only diff plus an empty-frontmatter changeset; no runtime code changed. Note for the PM: the issue arrived with assignees empty, as expected - the seat never writes that field.",
 "tests": "All at final commit df2db0de0 with a clean tree. Base green: `Tests 5 passed (5)`. Flake reproduced on base with a 50ms schema delay: `Tests 2 failed | 3 passed (5)` at :123 and :153. Fixed spelling under the same 50ms probe: `Tests 5 passed (5)`; under 1500ms: `5 passed` with 3.15s test time. Subject ablation (withheldFromServerSort -> false, mutation proved on disk, restored from HEAD with blob-hash equality): `Tests 2 failed | 3 passed (5)` at :164 and :197. Vacuity ablation (button row disabled): fails at the positive control :193. Package-wide: `Test Files 118 passed (118)`, `Tests 1089 passed (1089)`. Package type-check and lint pass (0 errors).",
 "mcp_calls": "1 - a single targeted search_issues, which answered 'API rate limit already exceeded'. Everything else went through repo-scoped REST (HTTP 200/201) and git. REST /search/issues answered 403, so the dedupe for the finding below is a title-level scan over all 273 open issues via the REST list endpoint; that bound is stated in the filed card.",
 "out_of_scope_findings": [
 "filed as #7075: four more plugin-detail test files (sortabilitySignal, lookupCellLink, headerSort, relationalSort - 17 tests) read a schema-derived verdict through a gate that only settles the fetch chain; measured by deferring the schema fetch 50ms inside the component (118 files: 4 failed / 114 passed), with this card's fixed file staying green under the same probe as the hot control"
 ]
}

---

Generated by Claude Code