I would say any evidence that there was an over-reliance on AI and no double checking by a real person/people whether predictions were likely to be accurate. Basically any overt lack of oversight at any point that could have contributed or blind faith in an AI system. I know this would still be debatable.

I think it's likely to contribute to one, but just like every other crash I expect people to debate over "what really caused it" incessantly. What would be the criteria for success here?

No, a flash crash wouldn’t count. By major I would say something like a 50% drop or more in several major tech companies - or a drop that the US government would consider bailing out due to its perceived degree of disruption.

would the flash crash count if it happend again?

How major?