Free during beta $19.99/mo at launch Get your free beta key →
Verdict Strategy BuilderNT8

How the engine works

Validation settings

The Quality tab configures "the validation gauntlet the top finalists are graded on — walk-forward folds, Monte Carlo, and a param-stress sweep." Blank fields use the engine default shown in each placeholder. For what the gauntlet's results mean, see Understanding validation.

Walk-forward mode

Mode Behavior
Rolling (default) Tiles many in-sample/out-of-sample folds across the span — the more rigorous default.
Anchored (proportional split) One proportional in-sample → out-of-sample → incubation split, so the equity curve colors as three contiguous blocks that scale with the dataset. Easier to read, less rigorous.

The knobs

Setting Meaning
Finalists validated (top-K) How many leaderboard finalists get the full gauntlet; the rest skip validation and show no verdict.
WF in-sample days Walk-forward in-sample window length, in days. Blank auto-sizes to the dataset span.
WF out-of-sample days Out-of-sample window length, in days. Blank auto-sizes.
WF step days How far each fold rolls forward, in days. Blank auto-sizes.
Monte Carlo iterations Trade-order permutations per finalist; more = smoother distribution, slower run.
Param-stress variations Jittered evaluations per finalist (±25% on every numeric parameter).

The fold preview

Below the knobs, a live preview shows the exact fold layout for the selected dataset — e.g.:

8 folds · IS 120d / OOS 30d · step 30d · incubation 45d · auto-sized to the dataset span

or, in Anchored mode:

Anchored split · IS 240d / OOS 60d · incubation 30d

If no dataset is selected yet, the preview says "Pick a dataset to preview the walk-forward fold layout." Warnings appear here too — for example when the dataset is too short for the requested fold sizes.

Sizing guidance

  • Leave the day-counts blank first. Auto-sizing adapts folds to whatever data you have; hand-tuned fold sizes are a refinement, not a starting point.
  • Top-K trades depth for speed. Validating more finalists costs time (each one gets full walk-forward + Monte Carlo + param-stress treatment). The default is a sensible balance; raise it when you plan to compare many candidates side by side.
  • Style presets set these for you — a Scalper preset uses short folds, a Swing preset long ones. Picking a style repopulates this tab.