Validation

Published walk-forward tests, historical simulations and observed live results are presented separately. Each engine report explains its test periods, weaker results, production configuration and remaining limitations.

Walk-Forward Analysis Overview

Walk-Forward Analysis (WFA) is a sequential validation protocol: optimise on in-sample (IS) data, then test on unseen out-of-sample (OOS) segments. Each slice reports performance on data not used by the optimiser.

Chart construction varies by study. Each chart identifies its historical sources and any transition into observed live returns. Overlapping OOS slices are reported separately; full-window reruns are historical research, not an untouched OOS record.

The reports document:

  • IS and OOS test periods
  • Overlap between slice periods
  • Profit factors and drawdowns, including weaker periods
  • Differences between balance and equity risk
  • The production configuration selected
  • Simulation assumptions and comparison limits

Historical tests inform development, but do not independently establish the absence of overfitting or guarantee future performance.

Archive · Production Baseline on 31 December 2025

Engine Status (as of 31 Dec 2025)

EngineStatusWindowParams
Engine ALiveFull-window (2014–2025)Frozen
Engine BLiveB5 sliceFrozen

This archive records the A and B production configurations at year-end 2025. Their strategy parameters remain frozen; operational fixes and protective actions are documented separately.

Forward tracking records subsequent live results, execution differences and documented incidents. Historical paths provide comparison context rather than predicted drawdown bands.

Further research is kept separate from the frozen live versions. Current results and monitoring notes are published on the forward-tracking page.

Forward Performance Tracking →