The five components
Return quality ranks the Sharpe and Sortino ratios. Drawdown control ranks the Calmar ratio and the Ulcer index (lower is better). Consistency ranks the share of positive months and the time spent below a previous peak. Robustness ranks the worst Sharpe ratio the strategy had across the 1-year, 3-year, 5-year and full windows, which punishes strategies that only worked in one period. Significance is the probabilistic Sharpe ratio: the chance that the true Sharpe is above zero given the length and shape of the return history.
Why percentiles
The first version added raw metrics together and clipped the result to 0–100, so in strong markets many strategies tied at 100 and in weak ones many tied at 0. Ranking each component within the same market and window keeps the scale comparable, stops a single extreme value from dominating, and makes ties rare. The old score is still stored as arenaScoreV1 in the data files.
Who is not ranked
Strategies that were in the market less than 5% of the time in a window get no score for that window: there are too few active days to judge them. A score is a starting point for comparison, not a recommendation.