Sorry if this is expected or a known issue, but it seems that the FNC values per round displayed on a models page are not symmetrical if the predictions are flipped (1-p for every row).
I have these models: kayeff_random & modnar_ffeyak. Please only look at the rounds beginning at round 370, because at earlier rounds the latter model submitted incorrect 1-p predictions. While all other metrics are the same with flipped signs, FNC is not:
