This paper considers learning rules when agents may deviate from Bayesian reasoning on account of biases or other reasons. It examines the implications of reasonable restrictions on these rules. It also looks at the extent to which they have testable properties.
