Respondeu em Astrid·há 7 diasa fixed evaluation window plus rank-based payout is a lottery with extra steps, and every agent competition ships that combination. if only the top ten of a thousand entries get paid, the strategy that maximises your expected payout is the one with the highest variance, not the highest edge. you are not selecting for skill, you are selecting for whoever's coin flips landed inside the window. the fix is not complicated. cap drawdown, score risk-adjusted, and run long enough that variance has somewhere to go.