I tried it, but it does not seem to change anything.

What is really strange is that it is as if there were a memory effect.

I modified the threshold on randvalue from 4 to 90, so that the YES case is more frequent, and it is easier to get it. And then I went back to 4. The first time I executed the survey just after going back to 4 gave me a randvalue of 91. In the screen, when executing the survey, it was displayed "NO", taking into account the modified threshold, whereas in the LimeSurvey database, I have a "YES" (see attached) as if it still takes into account the previous 90 threshold, and the following calculations differ from those I had on the screen.

I am doing the modifications on the activated version of the survey. Does this make any difference?

As Joffm says, please send the whole survey file then.

Because if it displays correctly during the survey, and later changes, this sounds strange. Would be interesting if there are other equations that might impact this.

Do you have other equations that include "DIS"? For example, if you check if DIS is "YES" or "NO" but use a single "=" instead of "==", then you might overwrite the actual value with the "compared" value.

Sounds the only reasonable explanation to me.

I agree with you but can't find any DIS='YES' with only one = in my code.

I attach here what my code using DIS is (it is just part of the code, so some variables such as DECISION 1 etc. are not defined in it). I am not using DIS anywhere else in the survey.