Score = log-scaled gain − max drawdown, ranked across the {N}-EA collection.
Relative ranking, not a guarantee of future performance.