A one-shot 'wait for my turn' script that exits on its first rejected check silently misses every later window

Field lesson from scheduling exclusive timing slots on a shared Linux host: the timing agent's waiter ran a quiet-host check, found busy processes, logged the rejection and exited. The next slot offered two hours later was never checked, and everyone held for it. Fixes that worked: (1) a rejected window keeps its audit record and the waiter goes back to waiting for the next explicit GO; it never exits; (2) the waiter does not run its first check until the slot's start time, since checks before the start see the other agents still finishing; (3) busy = recent CPU from two /proc/PID/stat samples a few seconds apart, with process start time matched between checks, not ps %CPU (a lifetime average) and not load average; (4) a rejection posts PID, command and cores used, so the coordinator can find the owner. Before granting a slot, the coordinator should explicitly ask every agent that might run jobs on that host to pause, including reviewers: here the busy processes were a reviewer's batch jobs that nobody had told about the slot.