Interpretation
How to read what each stage shows.
Feasibility
| Result | Read it as | Next step |
|---|---|---|
| Zero trades | Strategy didn't trigger, wrong data, or overly-restrictive filters | Fix before optimizing |
| Huge profit, tiny drawdown | Often a sign something's mis-set — costs, MaxBarsBack, or the date range | Double-check settings before trusting it |
| Small loss or flat on one symbol, majority still Pass | Not automatically disqualifying — a single weak symbol does not sink the batch | Continue to Optimization |
| Overall Pass, sum Net Profit/Drawdown look reasonable | Wiring looks sound across the tested symbols | Continue to Optimization |
Optimization
Optimization finds candidates — it doesn't prove an edge exists. Look for several neighboring parameter sets performing well rather than one isolated spike (the Robustness tab shows this directly); results that hold up across more than one symbol or interval; and a trade count large enough that the numbers mean something (a rule of thumb, not a hard line: fewer than 50–100 trades over your test window is thin). Be wary of a headline Net Profit that's high only because Avg Trade is large before costs and tiny after them.
Candidates and Monte Carlo
Prefer a steady, diagonal equity curve over a staircase carried by a handful of trades, and over a curve that flattens toward the end of the test period. Look at drawdown duration, not just depth. If Monte Carlo's median outcome already looks marginal, the backtest's best-case number is misleading you. If Batch Validation flags a candidate as failing, its real-world result likely depended more on favorable trade ordering than genuine edge. Use Deep Analysis, Time, and Risk alongside the core Analytics view before settling on a candidate — a strategy that only works in one seasonal window, or one that's fragile to a couple extra ticks of slippage, is worth knowing about before you commit further time to it.
WFO Config Matrix
You're looking for a robust region, not the single brightest cell. A green cluster spanning several neighboring Segment/OOS% values, with different segments picking different winning parameter sets, is a far stronger signal than one standout cell surrounded by blanks — that pattern usually means the result only works at one specific, arbitrary setting.
Validation
| Outcome | Read it as | Decision |
|---|---|---|
| WFO tracks or beats Initial | Strong generalization | Continue to Deployment Code |
| WFO profitable but noticeably below Initial | Some degradation is normal | Usually still workable — check the metrics table for exactly how much |
| WFO flat | Edge may not survive contact with unseen data | Rework or reject |
| WFO loses money | The walk-forward test failed | Do not proceed to Deployment Code |
| Initial baseline hidden with a warning | The out-of-sample windows could not be established, so the Initial comparison is hidden | Run WFO Config for this strategy first |