I'm realizing there should also be something in there about the number of crimes which an unconvicted criminal is likely to commit. Maybe their close call (being put on trial but not convicted) causes them to walk the straight and narrow, at least for a while. I am starting to think that a Bayesian model of this is what is called for. We do have some data (from DNA exoneration of the convicted, and also from clearly guilty who are not convicted because evidence was obtained improperly) as to how often we fail to convict the guilty, or convict the innocent, which we could plug into this.

Which brings up a larger point: what is N, really? Like, now, in the real world in my country or state or city, what is N? It would be interesting to try to estimate it.