A pull's check that fails while the check machine's disk is under its free floor is the machine's red, not the pull's: it is neither reworked nor parked #5455

A check that fails while the check machine's disk is under DISK_FREE_MIN is now treated as the machine's red: the crew doesn't rework or park it, and re-runs it once the disk recovers. bun x tsc --noEmit is clean. The new test, shared-red, crew-shared-red, and the vitals and checks suites pass. I did not run the whole suite, so the forge's check will be the first full run.

• Recording the reading: every red check now records the disk's free fraction on its run, as an optional diskFree on CheckDetail (src/core/runs.ts), set in src/checks/service.ts. A red that named no test gets an empty detail carrying just the reading. Green checks are untouched, and there's no new constant or migration.
• Waiting and re-running: waitOnMain in src/crew/dispatch.ts checks the disk first, beside sharedWithPulls. A pull under the floor is held in memory with no rework, park or strike. The crew tick calls recheckForDisk, which re-runs the held pulls once the disk is back above the floor, if the head hasn't moved. Like the existing waiting list, it is forgotten on restart; a push or a person's ask re-checks the pull.
• The vitals card: src/kernel/vitals.ts shows one card, "The check machine's disk was full when N checks failed: X% free, under the 10% floor…", covering the last hour of such reds. It uses the lowest percentage recorded. The logic lives in a new src/checks/disk_red.ts.
• Tests: test/checks-disk-red.test.ts injects the disk reading as plain data. Its three cases cover the three requirements you listed: a red under the floor is held and re-run only after recovery, with one card for three checks; a red above the floor parks as before; the service records the reading. I wrote them to fail without the change but did not run them against the old code.
• Existing tests changed: two in test/checks.test.ts asserted the exact checkDetail, which now carries diskFree, so I loosened them. test/vitals.test.ts counted queries; the card adds one, so it now expects 10, not 9. These tests were right about the old shape, not wrong.
• Studio words: src/web/studio/words.ts skips the "0 passed, 0 failed:" note when a red names no test.

Left out: why the disk fills, any disk-aware change to the green path, and persisting the wait across restarts.

To try it: bun test test/checks-disk-red.test.ts test/shared-red.test.ts.

Closes #5453

A check that fails while the check machine's disk is under DISK_FREE_MIN is now treated as the machine's red: the crew doesn't rework or park it, and re-runs it once the disk recovers. bun x tsc --noEmit is clean. The new test, shared-red, crew-shared-red, and the vitals and checks suites pass. I did not run the whole suite, so the forge's check will be the first full run.

• Recording the reading: every red check now records the disk's free fraction on its run, as an optional diskFree on CheckDetail (src/core/runs.ts), set in src/checks/service.ts. A red that named no test gets an empty detail carrying just the reading. Green checks are untouched, and there's no new constant or migration.
• Waiting and re-running: waitOnMain in src/crew/dispatch.ts checks the disk first, beside sharedWithPulls. A pull under the floor is held in memory with no rework, park or strike. The crew tick calls recheckForDisk, which re-runs the held pulls once the disk is back above the floor, if the head hasn't moved. Like the existing waiting list, it is forgotten on restart; a push or a person's ask re-checks the pull.
• The vitals card: src/kernel/vitals.ts shows one card, "The check machine's disk was full when N checks failed: X% free, under the 10% floor…", covering the last hour of such reds. It uses the lowest percentage recorded. The logic lives in a new src/checks/disk_red.ts.
• Tests: test/checks-disk-red.test.ts injects the disk reading as plain data. Its three cases cover the three requirements you listed: a red under the floor is held and re-run only after recovery, with one card for three checks; a red above the floor parks as before; the service records the reading. I wrote them to fail without the change but did not run them against the old code.
• Existing tests changed: two in test/checks.test.ts asserted the exact checkDetail, which now carries diskFree, so I loosened them. test/vitals.test.ts counted queries; the card adds one, so it now expects 10, not 9. These tests were right about the old shape, not wrong.
• Studio words: src/web/studio/words.ts skips the "0 passed, 0 failed:" note when a red names no test.

Left out: why the disk fills, any disk-aware change to the green path, and persisting the wait across restarts.

To try it: bun test test/checks-disk-red.test.ts test/shared-red.test.ts.

The full gate on land was red; the pull is open again. Red tests:

• apps: through a rollout > the forge's rollout: the new forge is ready only once the app serves, and a socket reconnects to it in under a second with no 502
• apps: through a rollout > the forge's rollout: the new forge is ready only once the app serves, and a socket reconnects to it in under a second with no 502 [789.54ms] [shard 68]