Survey Cut-off Value

I am also conducting research on the effect of High-Definition transcranial direct stimulation (HD-tDCS) on implicit motor sequence learning, via SRTT (serial reaction time task). 20 participants receive 30min stimulation, 20 receive 15min, and 20 are the control group (sham) and receive no stimulation.
In order to exclude the data of participants whose reaction times may have been affected by the stimulation (discomfort, itching, distraction, slight burning, etc.), participants must fill out an "unpleasantness" survey, consisting of several questions to their perceived unpleasantness.
My question is: How can I empirically/ statistically choose a cut-off value, in order to determine from which point on, data has to be excluded?
Thank you so much in advance for ideas and answers.
I'd start by making lists of everything, survey stuff and results stuff.
No I wouldn't.
I'd realize that with n = 20, I'd never reach a reasonable conclusion about data exclusion.
Unless there was/were a/some cliff-like data shape. Ex; Everybody who answered only 3 of 7 survey questions had no reaction to treatment.
No, n is too small. of 20 you'll have a variety of survey results and a variety of treatment results, so n per combination is gonna be very small, some zeros.
And 20 in the control group will have some who react to (no) treatment and some who didn't react but were affected.
There are at least 60 combinations, 3 n = 60, too many combinations for the n.
Start at go, 3 groups, who responded how much to treatment. Go from there.
We can precisely analyze small n situations, but we know that we're only kidding. Ourselves or others.