[HugoXK]

Fixes #1843

Written-by: Dev3
Owner: Dev3
Reviewer: Reviewer
Claim-Id: bbcd50297ba84d3dad4f26da8e7b5dae
Branch: dev/3-flaky-panel-state
Base: 28e887b
Docs-not-needed: test-only change to one timing assertion and its whole-file mirror; no owned path or behaviour changes.

What was wrong

core/test/project-board-panels.test.js was order-dependent and failed at random on merge groups — it ejected #1825 once: run 41046947850 red, run 41047579302 green at the same tip 33a5d463. Both reds named the same assertion:

✖ the settings panel memory save lands in the Project memory through the real endpoint
AssertionError: the panel reports the state the host answered
+ actual:   'Assistant settings saved.'
- expected: 'Memory saved.'

Two defects, both in the test harness, no production code involved:

• The hook slots are module-level (core/test/project-board-panels.test.js:33-34) and render() resets only the cursor (:58), never the slots. A value one test clicked therefore stayed readable by the next test — the bind/assistant test leaves 'Assistant settings saved.' in the status slot.
• settle() was a fixed ten event-loop ticks (:77), not a wait for the host's answer. On a loaded runner the memory save had not landed when the assertion re-rendered, so it read the stale line above.

What changed

• Every test starts from empty slots (resetSlots()), so nothing one test clicks can be another test's value.
• The three status assertions poll the text the panel draws (waitFor / statusAfter, 100 polls of 25 ms, bounded) instead of waiting a number of ticks. A timeout still fails on the value it reads, as the CI failure does above. The one-tick read in the last refusal test is the same defect class, so it uses the primitive too.
• The primitive is the one core/test/retired-bot-paths.test.js adopted in test(retired-bot-paths): wait for both reviews instead of fixed settles (#1833) #1834, kept local to this file (no shared helper). test(retired-bot-paths): wait for both reviews instead of fixed settles (#1833) #1834 had no cross-test hook-slot sharing to fix, so only the wait applies there; the per-test slot reset is specific to this file.
• core/test/project-board-panels.test.js and patches/ds-bot/files/test/project-board-panels.test.js are byte-identical (sha256 521a123d269a73decae6b026202bfbc288e02c1803e803f882ebebcac170e6e8, 14,725 bytes each). retired-bot-paths.test.js is untouched.

Red first, on base 28e887b8

The CI order (bind/assistant test first, memory save after it) run twice in one process on an idle machine does not reproduce it: 12/12 green both times. On an idle runner the ten ticks suffice, which is why this only reds merge groups.

The red needs the load, so I reproduced it with the condition that made it flaky: a scratch copy of the base file with a 120 ms delay on project-memory-write.

not ok 3 - the settings panel memory save lands in the Project memory through the real endpoint
  expected: 'Memory saved.'
  actual:   'Assistant settings saved.'
# tests 6  # pass 5  # fail 1

That is deterministic, and the order dependence shows in the actual value, not only the timing: with the same delay, moving the memory-save test before the bind/assistant test makes the value '' — the stale line is whatever the previous test left. Removing the wait entirely fails the same way, so the two defects are independently sufficient.

Verification at the tip ed4ddb9c73ad5e7de36f39fcbc0d1dcc6a8bc5dd

Ran:

• 20 consecutive runs of the file (node --import ./test/harness-resolve.mjs --test test/project-board-panels.test.js): 20/20 green, 6/6 tests each.
• Two blocks swapped: 5/5 green; swapped with the 120 ms delay: 3/3 green; delay alone: 3/3 green; the file run twice in one process: 3/3 green (12/12 tests).
• TZ=UTC ARCH_DOCS_BASE=28e887b8 npm run verify on the committed tree (git status --porcelain empty): exit 0 — core 2221/2221; core lint 72 warnings, 0 errors; plugin lints 36 / 9 / 2 / 16 warnings, 0 errors; arch-docs: OK (2 changed file(s)); build-client --check green; ds-bot-patches check 10 patches apply cleanly in order and core/ equals ds-bot/ + patches/ds-bot/; plugins social 541, tree 127, timeline 135, workflow 294; .github/scripts 549; interop 37, live-e2e 24 — 0 failures everywhere.

Not run:

• CI tests-ok at this tip — that is the check this PR exists to stop flaking; it must be green before queueing.
• The smoke test: no index.js or src/host/ change.

Checkpoint

{
  "pr": 1850,
  "claim": "Dev3",
  "issue": 1843,
  "summary": "The panel test file kept hook slots across tests and waited a fixed ten ticks, so a loaded runner read the previous test's status line. Each test now starts from empty slots, and the status assertions poll the drawn text with a bounded wait. The wait primitive stays local to this file and mirrors the one #1834 adopted; nothing in production code changes. The file is green 20 runs in a row and with the two blocks swapped, and under the 120 ms delay that reds the old file.",
  "files": [
    "core/test/project-board-panels.test.js",
    "patches/ds-bot/files/test/project-board-panels.test.js"
  ],
  "checks": {
    "ran": "20 consecutive runs; swapped-order runs; 120 ms delay probes; TZ=UTC ARCH_DOCS_BASE=28e887b8 npm run verify; ds-bot-patches check; build-client --check",
    "result": "green",
    "at": "ed4ddb9c73ad5e7de36f39fcbc0d1dcc6a8bc5dd"
  },
  "next": "none — ready for review"
}

[Birfy]

Verdict: PASS at ed4ddb9
Written-by: Reviewer
Attestation: ed4ddb9 role=Reviewer

Ran at this tip:

$ cd p1850/tip/core && TZ=UTC node --import ./test/harness-resolve.mjs --test test/project-board-panels.test.js
# 20 consecutive runs, 6/6 each: tests 120  pass 120  fail 0
$ TZ=UTC node --import ./test/harness-resolve.mjs --test test/pb1850b.test.js
# base blob of the same file + identical 120 ms delay on project-memory-write: tests 6  pass 5  fail 1
$ TZ=UTC node --import ./test/harness-resolve.mjs --test test/pb1850t.test.js
# tip blob + the same 120 ms probe: tests 6  pass 6  fail 0

Red first. The base blob of core/test/project-board-panels.test.js at 28e887b8 is green on an idle box (that is why this only reds merge groups), so the load the failure needs was made explicit rather than waited for: one 120 ms delay on project-memory-write, applied identically to a scratch copy of the base blob and of the tip blob (both copies deleted after the run, nothing in the change set touched). Base fails exactly as CI did — expected 'Memory saved.' / actual 'Assistant settings saved.' — and the actual value is the previous test's line, which is the module-level hook slot the PR names; tip passes under the same probe. Tip is also 20/20 green in a row unmodified.

Scope. compare 28e887b8...ed4ddb9c = 1 commit, 2 files, both test-only: core/test/project-board-panels.test.js and its whole-file mirror patches/ds-bot/files/test/project-board-panels.test.js, sha256 521a123d…0e6e8 on both, 14,725 bytes each. No production file, no owned path; Docs-not-needed: is in the body and in the tip commit message.

Tier. Risk gate reported T1 on this tip (risk tier (report only) job 124149100556, risk-gate (report only): PR #1850 tier T1), so no PASS is required; this one is written anyway because the round asked for a verdict, and it is the flake class #1843 records.

Checks at the tip. 14 check runs, all completed/success, 0 failures: tests-ok, merge-gate, enqueue-guard, reviewed tip, risk tier, agent-audience ×2, judge, plugins (tree|social|timeline|workflow), ubuntu-latest · Node 24 · DSH 0.2.0-rc.2, package. Mergeable clean, not draft. The PR body carries a valid ## Checkpoint block (1034 characters, pr 1850, claim Dev3, issue 1843, two files from this diff).

Not run: the smoke test (no index.js / src/host/ change) and a full npm run verify locally — the tip's own tests-ok covers the whole suite, and this diff cannot reach anything but the one test file.