Yeah, haha. I have tried so many supplements and continued testing for extended periods of time just to see if it was a placebo. Gwern takes it really next level with blind-testing a lot of supplements, but I found that a bit too difficult.
One of the features that I did to help reduce that is rolling aggregate correlations, so for instance, compare a full week's worth of productivity, sleep and supplements to a week without that supplement. That helps average out a LOT of daily variance.
At the 21 days aggregation - the data starts getting noisy again. Human habits don't want to be a linear regression, ie. baselines of productivity and sleep drift over time, regardless of supplements and habits.
PS - See that you own https://zenobase.com/, really dig how many integrations you've done with it! It's a great site.