DROP ALERT NOTES
Log every alert decision, not just every alert
The alert log tells you what arrived; the decision log tells you what your rules did about it. The second log is the one that tunes the system.
Most alert setups log what arrived: the alert, maybe the message that went out. That log can tell you the stream was alive, but it cannot answer the question that matters after a month: are my rules any good? The answer lives in a second log, one line per decision, that records what your rules did with each alert and why.
What a decision record holds
One compact line per alert, written whether the alert passes or not: the alert's id, the rule that judged it, the verdict, and the reason in one word. The dedupe guide's keys make the identity side easy: the alert key tells you it is the same alert, and your verdict tells you what your pipeline chose. A decision line looks like: id, rule, pass or drop, reason, and the fair price if it mattered.
The reason word is the whole trick. Threshold-fail, duplicate, cooldown, level-filter, passed: five reasons cover almost every pipeline, and they turn a log into a report.
What the decision log answers
After a week, group by reason. A flood of threshold-fail means your stream threshold is generous and your code does the filtering, which is fine but worth knowing. A flood of duplicate means the same move alerts repeatedly and your dedupe is earning its keep. A rule with zero passes in a week is dead weight or broken, and the log is the only thing that can tell those apart.
The dedupe guide ends with the same posture: measure the effect. The decision log is the instrument that measures it.
The two rates to watch
Two ratios do most of the reading. The pass rate: decisions that pass, divided by alerts received, per rule. And the collapse rate: alerts in, divided by messages out. A healthy prematch pipeline might collapse twenty alerts into three messages; a healthy live pipeline fewer. When either ratio drifts, something changed, and the decision log tells you which rule moved.
Track them per sport and per phase, because a single blended number hides the one sport that is flooding you.
Keep it small
The decision log is one line per alert, a few dozen bytes, so it stays cheap even on a busy week. Write it beside the relay, rotate it weekly, and review it before every threshold change. The drop alerts guide covers choosing the thresholds; the decision log is how you keep them chosen well.
Arrivals tell you the stream worked. Decisions tell you the rules did. Log the second, and the first becomes evidence instead of noise.