I happen to think you’re right about Harman’s dead dictator case and I feel the same about the fake barn case. I believe I’ve started calling these Gettierish cases and if I’m not mistaken (highly unlikely) I think I picked this up from you.

The reason I don’t like the third definition is that I think there are some cases, like Harman’s dead dictator case, that feel like an importantly different kind of case to the ‘coincidental truth’ cases that I think are the core kind of Gettier case.

Perhaps you’re right about what’s crucial is that the beliefs are reasonable, whether or not they’re justified. I’ll have to think about that more.

I’d be happiest with the third, but there’s just one more wrinkle. There are some views on which a justified belief just is knowledge. I have in mind Sutton’s view and although some people don’t like to own up to this, it seems some people’s views concerning warranted assertion commit them to this view. As tempting as it might be to say such views are committed to the impossibility of Gettier cases, advocates of such views can say pretty much what the rest of us want to say about such cases. They can say a subject is justified in believing p, p is true, but the belief that p is not justified as it fails to constitute knowledge (drawing on Bach and Engel’s distinction between personal and doxastic justification). They can say as Sutton does that the belief in Gettier cases is reasonable but not justified. So, I’m not quite sure that the definition of a Gettier case should be one in which there is justified belief that fails to constitute knowledge.