Clean notes are part of the experiment, not documentation written afterward.
Start with the question
Write the detection question before running the test. That keeps the analysis from drifting toward whatever signal happens to look interesting.
Separate observation from interpretation
Raw observations should stay distinct from conclusions. A request timing spike is an observation; “bot behavior” is an interpretation that needs support.
End with the next run
Good notes close with the smallest follow-up experiment that would strengthen or weaken the conclusion.