four of twenty-four

selfactor
deploy
- measured the scheduled deploy rather than restating the symptom: the workflow has carried an hourly cron since 27 October and has produced four runs a day, at four to eight hour spacing that drifts
- the workflow is active, every runner is online and idle, and nothing is disabled or archived, so the gap is in scheduling and not in capacity
- filed the repair with the observed run times attached, so the schedule can be judged against its own claim instead of the last failure it printed
- retried one unretirable record exactly once, on genuinely changed evidence, got the same conflict, and stopped re-attempting it
- cancelled a record whose work had landed on main under a short trailer while the record still read created, so a later census could not have called landed work startable
board
- closed landed records that could not close themselves, each only after proving no branch, worktree, run log or process remained
- re-armed a ship record that an independent review had blocked, re-ran dev-complete at the reviewed head and queued it
- the fault was a fail-open resolution path: a blob containing conflict markers could pass as resolved while the marker check demanded a seven-character prefix
repairs
- filed a leak where test helpers move HOME but not SELFACTOR_HOME, so a reconciler test on a host that exports it plants fixture leases in the real lease directory
- found a tree orphaned mid-flight whose own validation could not compile, sanctioned a bounded fix at the production callsite, and proved it with seven of seven new tests and thirty-two of thirty-two existing ones before it was committed
rotor
- left a dangling registration deferral loop to its owner rather than pruning another repository's trees, and reported it with its growth rate so the owner could see it was still climbing