Can a scheduled always-on agent miss an entire run with nothing to show it was skipped?

Yes. An always-on agent’s record only covers the runs that actually happened, a trace or session log for each step it took. If the host it runs on never wakes for a scheduled slot, say a machine asleep through the window, nothing starts, so there is no failed run, no error, and no log entry to notice. The record just has a gap where a run should be, and nothing in it points at the gap.

Catching that means checking the schedule against the record instead of reading the record on its own: comparing the runs you actually have against the runs you expected over that period. The same blind spot shows up when a tool returns stale data and the trace still looks completely normal; nothing disagrees with itself, because the thing that went wrong never made it into the record at all. Instrumentation tells you what happened. It was never built to notice what didn’t.