3 ms·

I assume I’m not the first making this point, but I think they’re evaluating all the models, not just the ones that they know did well.