The phenomenon where a strategy's logic or parameters have been tuned so precisely to fit a specific stretch of historical data that backtest results look exceptionally strong — but that performance is really just coincidentally fitting past noise, and drops off sharply when applied to new data or a live environment.

An evaluation metric that weighs a strategy's return together with the volatility risk it took on to achieve that return, measuring how much actual return is earned per unit of risk taken — letting strategies with different volatility levels be fairly compared on the same basis, rather than judged on raw return figures alone.