Posterior distribution over a countable setMof continuous data-sampling distributions
piles up at

-projection of the true distribution

on

, provided that
the

-projection is unique. If there are several

-projections of

on

, then the
posterior probability splits among them equally.
Fulltext