Validation · 2026-08-12
Walk-forward is not optional
A strategy that only works where it was fitted has memorised, not learned. Twelve windows, judged only out of sample.
Every candidate we take seriously is fitted on one span of history and judged on the span it has never seen. Then the window rolls, and it happens again. Twelve times. The fit is discarded; the judgement is kept.
A single in-sample Sharpe is a costume. Walk-forward efficiency — out-of-sample edge as a fraction of the in-sample edge that promised it — is the number that decides whether the idea is allowed near capital. Our gate is 0.6. Most research dies here. That is the point.
The chart on Folio 02 is schematic, not a trophy. It is the shape of a process that refuses to flatter itself.