Can post-processing fits (and other operations) be stochastic?

[MichaelClerx]

Post-processed data can be used as input for e.g. fits.

From a reproducibility point of view, especially if we'd like to do post-processing on the fly, the results would always be the same.

Is this an issue? Should we try to avoid any randomness?

At the moment, the only stochasticity is in the time constant estimation, which uses repeated fits from random starting points. I'm not sure this is the best way to do it anyway, so will try to rewrite that. And if the output changes slightly between runs, that only affects plots - as we don't use time constants as input for fits.

But in general, should we allow / ban stochasticity?