A lot of experiment logs record what happened but skip why that result was plausible. That missing context is where reproducibility usually breaks.
I now keep each run in three blocks:
- Process settings and constraints
- Live observations during the run
- Immediate interpretation and next check
This format makes it much easier to revisit a dataset weeks later and recover the reasoning behind each parameter change.
If future me cannot repeat the run from the note alone, the note is incomplete.