Thank you @einat.orr for your thoughts! Yes, the approximate reproducibility is useful, we do scientific analysis so being able to reproduce historical results - exactly, or approximately - is very good. I suppose I could also write a separate "redactions" log which could be referenced as necessary.