The fundamental problem with false convictions is that they are unobserved, and in general, unobservable. We don't spot them when they happen-if we did, they wouldn't happen-and in most cases we can't identify them after the fact. We have no general reliable test for innocence or guilt; if we did, we'd use it at trial. As result, we often say that we don't know for sure whether a convicted criminal defendant is innocent or guilty, or even that we can't know for sure. But this isn't exactly true-or rather, its truth depends on who we mean by "we."
Gross, Samuel R. "Pretrial Incentives, Post-Conviction Review, and Sorting Criminal Prosecutions by Guilt or Innocence." N. Y. L. Sch. L. Rev. 56, no. 3 (2011-2012): 1009-30.