A test file that fails in its shard is run once more alone; one that passes alone is named as 'fails only beside others' and does not fail the pull #2416

Six issues were parked twice on 16 Oct, and most of their reds were tests the pull never touched: 'the pages', 'token counts on the pages', 'fork at the origin…', 'a deck chooses among sibling pulls', 'variant votes … terminal', 'a green agent pull and its reviewer'. Each of those pulls added a file to GATE, which reshuffles which files share one bun test process (scripts/shards.ts). A test that leaks state into another then fails somewhere new each time. #2270 does not catch this, because no single test fails on 3+ pulls.

Done looks like:

• [ ] In scripts/test-shards.ts, after a shard goes red, each failing file is run once more as its own bun test. If it fails alone, the check is red, as today. If it passes alone, that file does not fail the check, and the tail says fails only beside others: <file> (shard-mates: …) so a person can find the leak.
• [ ] A file is retried at most once, and at most 3 files per shard; beyond that the shard stays red.
• [ ] Test with two fixture files: one sets a global that breaks the other only when they share a process. The run is green and names the file. A fixture that fails alone stays red.
• [ ] Run the test-shards tests and quote the result.

Parked by the crew: the check failed ("toolbar actions > (unnamed)") and the decider saw no point in another try (pull #2427 closed). Planner: split this smaller, or rewrite what done looks like; a person can also say what to try.

Parked by the crew: the check failed ("quotas in the terminal > (unnamed)") and the decider saw no point in another try (pull #2433 closed).

Parked twice since a person last spoke here, so it is the planner's now. Every reason:

1. the check failed ("toolbar actions > (unnamed)") and the decider saw no point in another try (pull #2427 closed)
2. the check failed ("quotas in the terminal > (unnamed)") and the decider saw no point in another try (pull #2433 closed)

The planner splits it, replaces it with a fresh issue, or defers it; a person can also say what to try. Nothing starts here until then, and the board lists it.

State: split into #2444 and #2445. #2444 is a pure besideOthers function in scripts/shards.ts with unit tests, and does not change how the gate runs. #2445 wires it into runShard with the two-fixture test. Both pulls were red on files they never touched (toolbar actions, quotas in the terminal), so #2445 says to name such a red in the pull rather than fix it there. #2417, #2420, #2423, #2424 and #2425 now wait for #2445.