numeraire.core.evaluators

Native evaluators (numpy/scipy, no heavy deps) — the performance family.

Evaluators dispatch by capability and emit rows of the tidy result schema, so the metric always matches the object (VoC’s headline is timing Sharpe, not R²). Each carries requires (the capabilities an OOS output must expose) and registers itself in the open evaluator registry so external packages add peers without editing core.

SharpeEvaluator	Annualized Sharpe ratio of the realized strategy returns (the timing headline).
MeanReturnEvaluator	Annualized mean of the realized strategy returns.
CEQEvaluator	DGU (2009) certainty-equivalent return of the realized strategy returns (economic value).
AlphaEvaluator	Time-series alpha of the strategy vs a factor benchmark (HAC t-stat).
StrategyReturnEvaluator	Per-period (time-indexed) realized strategy return — one result row per date.
OutOfSampleR2Evaluator	Out-of-sample R^2 of a forecast vs a benchmark, 1 - SSE_model / SSE_benchmark (percent).
SquaredErrorDiffEvaluator	Per-origin squared-error difference (benchmark minus model), one row per date.
ClarkWestEvaluator	Clark-West (2007) MSPE-adjusted test of the forecast against its nested benchmark.
CrossSectionalR2Evaluator	Cross-sectional R^2 of mean realized returns on mean predicted expected returns (OLS).
AverageAbsAlphaEvaluator	Average absolute pricing error (mean over assets of |mean realized - mean predicted|).