Report comment

There are a couple reasons for this : Neural nets usually haven't got a lot lop-sided data so can have bizarre noisy evals for these positions, and they're normally a lot slower to evaluate than a hard-coded heuristic.