A main form of insight is a hypothesis that one hadn't previously entertained, but should be assigned a non-negligible prior probability.

I think of this as P(hypothesis H is true | H is represented in my mind) > P(H is true | H is not represented in my mind), largely because someone likely did some calculations to hypothesise H (no matter how silly H may seem, e.g. "goddidit", it's better than a random generator, with few exceptions).

So, in a way, I consider the act of insight as evidence (likelihood ratio > 1) for the insight itself (the hypothesis).