Make the hypothesis specific
Name the behaviour you are trying to measure before choosing parameters. It is easier to challenge a precise explanation than a collection of attractive metrics.
Separate discovery from evaluation
Use one part of your data to explore and another to evaluate the idea. Keep track of how many alternatives you tried; a result selected from many experiments needs that context.
Write down the next test
A useful research note ends with a concrete question. Describe the observation that would strengthen or weaken your explanation.
The next test should teach you something, even when the result disappoints.