Issues with Non-Accidental
Properties
Is it just Bayesian inference?
Then why not model all information?
This may fit Lowe
Biederman relies more on certain inference.
See also Feldman, Jepson, Richards.