3. Metrics

July 26, 2026 · View on GitHub

Status: Draft — content TBD.

This section will catalogue the scalar and time-series fields that a producer MAY populate on:

BacktestMetrics — per-run metrics (see Run.backtest_metrics).
Summary — per-engine-slot aggregate metrics (see EngineSlot.summary).
MonteCarloTest — per-slot significance-test results (see EngineSlot.monte_carlo_tests).

TODO

Full per-run BacktestMetrics field catalog: capital / P&L, return curves, risk-adjusted (Sharpe / Sortino / Calmar / profit factor / annual volatility / VaR / CVaR), drawdowns, TWR (alpha-only) variants, trade statistics, trade durations, trade returns (averages, medians, current-only variants), frequency, exposure, best/worst, period win-rates.
Full pooled Summary field catalog, including cross-run consistency and in-run stability metrics.
MonteCarloTest structure: name, method, metric, observed_value, null_distribution, n_permutations, p_value, alternative, seed, window, universe_key.
Reserved MonteCarloTest.method values (shuffle_ohlcv, and reservations for future flavours).
Semantics for each metric (definition, units, expected sign, boundary cases such as zero-trade runs).
Which fields SHOULD be extracted to blobs (see 4. Blobs).
Handling of unpopulated / nil / NaN scalar metrics.

Until this section is filled in, refer to the reference implementation's design doc for the canonical field list: