got a weekly server scan that's supposed to only alert on changes, but it kept resending the same near-identical pile of findings every sunday. felt like a repeat of last week because it basically was. the job keeps a state file so it knows what changed, except half the things it compared never stopped changing, so "changed" was always true.
the journal check keyed on the exact file size string so it always changed, because the size always grew a little. the memory check keyed on the ordered list of top processes, and two near-identical python processes swapped places every week. the network check keyed on the exact set of bound ports, and a container churned a port or two each time. three keys that could never match their own last value.
fixed all three to be sticky and threshold based. memory only fires if real usage crosses a hard line. the network only fires on a first sighting or when several ports appear or disappear at once, single-port churn stays silent. the journal alert switches on once near the cap and only clears after the size drops well below it. also found two state flags that silently dropped out of the file on every rewrite, which is why my fix kept breaking the next week. now everything persists every run.
verified it runs clean and stays silent back to back. then got the second job fixed the same way: the daily screenshot digest was dumping its whole internal breakdown into the message instead of the one casual line i actually want, because the prompt it ran on was literally truncated mid-sentence in the stored config. rewrote it with a hard output contract. both jobs shut up now unless something's real, which is the whole point of scheduling them.