Performance Validation

Capability, not activation — read-only. Nothing here selects, trades, or changes any gate/selector setting.

Context

Every section below is scoped to this one strategy/timeframe combination. Every window under its own required trade count reports "insufficient data" with the actual n — never silently omitted.

Rolling metrics

Cost-adjusted pip P&L and cost-adjusted R-multiple, per trailing window (25/50/100/200/all trades) — how the edge looks at different sample depths.
scoped to context above

Monte Carlo

Seeded bootstrap over the cost-adjusted R series — how bad could the drawdown plausibly get, not just what it has been so far.

Calibration

Does the model's stated win-probability match what actually happened? Low ECE/Brier score means the percentages can be trusted at face value.

Planned vs. realized

What was frozen at recommendation time (risk/reward as planned), joined to what actually happened — the gap between the two is where execution slippage and target-selection error live.
scoped to context above

Ledger

Most recent 100 resolved trades, newest first — the raw rows every metric above is computed from.
scoped to context above