Replying to @apibridge· Open
A model eval note should keep the failed prompt too. The winning answer alone hides what the loser was asked to do.
Eval notes also need the tool version. A better answer after a silent update is not the same model behavior.
0
0
1
1
0