What broke
Failures with the measurement attached: what I believed, what was true, and the cheap check that would have caught it.
-
6–9 minutesThe messenger died with the loop
A nightly loop was dead for nine nights, and nobody knew — because the thing that should have told me died on the same fault.
-
8–12 minutesOne rule, four copies, and only one had the test switch
The question of when the next press release was due got answered in four separate places — and only one of them could be tested.
-
7–11 minutesThe charset guard went green for months on unchecked files
A guard that checked character encoding passed for months — because the rule never reached the source code it claimed to cover.
-
11–16 minutes147 runs, 100.13 dollars, zero posts
The most expensive agent loop in my house ran 147 times in 30 days and never delivered a single post. The text was fine. The carrier was not.
-
5–8 minutesA watcher reported zero, and zero was the wrong answer
My X auto-publish gate checked a catch block, not the response — so a 401 and 'nothing to post' both read as null.
-
3–5 minutesThe chain that says sent, and sends nothing
The card in my cockpit says the video went out. No video has ever gone out. This one is not fixed yet.
-
3–5 minutesEvery failed job left a browser running
The machine got slower every day and the logs were clean. Failures were leaving something behind, and nothing counts what should have ended.
-
9–13 minutes
Four ways a system tells you it is fine
Six faults across six systems, none of which produced an error. In every case something I had built reported that things were fine.
-
4–7 minutesTwelve days of silence, every check green
A signup form stored nothing for twelve days. No error anywhere, every check green — and the strategy decision that was about to produce.
-
2–3 minutesThe watchdog that died quietly, and took the evidence with it
A guard killed the run that was carrying its own finding. Because it killed it, nobody noticed for nine nights.