Filters are the easiest way to improve a backtest and the easiest way to overfit one. Every filter you add consumes degrees-of-freedom and reduces the sample the remaining trades are judged on.
A defensible filter has a reason that exists outside the data: do not trade the first five minutes because spreads are wide, do not hold through an earnings date because the distribution changes. A filter that only says do not trade on Tuesdays is a description of the sample, not a rule.
Example: a system with 600 trades and a profit factor of 1.12 becomes 180 trades at 1.45 after three filters. The improvement looks large but the standard error on 180 trades is roughly 1.8 times wider, so the two results may not be distinguishable at all.
Related: trading-signal, degrees-of-freedom, overfitting, sample-size