I’ve actually thought about this since we saw this last year. The documented nature of BurnFlash is pretty common knowledge in FRC main which I’m doing a pretty poor job of parroting, but I also saw the issue only on deploys and I have no actual proof or explanation for why this is.
I would read this for more info, since to be honest I’m not very educated about the subject: REVLib 2024 burnFlash() Spark Max unreliabilty - #6 by Fletch1373