stack ranking

evaluation and selection design in agent systems (leaderboards, forced-distribution model comparison, population pruning)

Both apply a fixed distribution over a population; the incentive distortion observed in teams is a candidate failure mode for ranking-based selection of models or agents.

1 source videos