Tag
Belote AI
OpenSpiel's Python ISMCTSBot ignored its seed
Two runs of the same seeded ISMCTSBot match made different decisions, because the default resampler never got the bot's random_state. Fixed upstream in open_spiel#1584.
Test an evaluation harness on random vs random first
With no skill gap between the two sides, the harness should report a margin indistinguishable from zero. If it can't show no difference, it can't be trusted to show a real one.