RESEARCH LAB / PHASE 02
Backtesting
Chronological walk-forward previews with separate training and validation windows.
Demo channel connecting
Research job previewOnly a fixed walk-forward report is available in Phase 2. No dataset is fetched and no backtest computation is launched.
Chronological validation plan
Deterministic synthetic sample · preserve time order and avoid future leakageWindow choices describe this fixed report; they are not engine parameters.
Walk-forward folds
EURUSD → GBPUSD · authored example, not research results| Fold | Training period | Validation period | Validation n | Net expectancy $ | Profit factor | Max drawdown $ | Review |
|---|---|---|---|---|---|---|---|
| 01 | Day 01–20 | Day 21–25 | 18 | +2.40 | 1.31 | 32.00 | Small out-of-sample window |
| 02 | Day 06–25 | Day 26–30 | 22 | −0.80 | 0.92 | 46.00 | Economic stability fails |
| 03 | Day 11–30 | Day 31–35 | 16 | +1.10 | 1.14 | 28.00 | Independent evidence needed |
Fixed report ready for inspectionNo optimisation or random train/test split
Freeze before validation
Parameters are chosen from each training window and fixed throughout the following validation period.
Cost sensitivity matters
Report net expectancy under wider spreads and conservative fills. Do not promote a strategy from win rate alone.
Watch regime dependence
Compare held-out sessions and regimes. Three authored report rows establish no market evidence.