Operations
Reading the gauntlet stages
When a run finishes evolving, its state flips to Validating and the top finalists enter the validation gauntlet. Instead of an opaque badge, the run card shows the gauntlet as visible stages, each with its own progress:
| Stage | What's happening |
|---|---|
| Walk-forward | Each finalist is re-run across the rolling fold plan (the stage counts folds). |
| Monte Carlo | Trade sequences are permuted with random slippage into a drawdown distribution. |
| Param stress | Parameters are jittered and the strategy re-evaluated for plateau vs. cliff behavior. |
| HBT recon | The in-house high-fidelity cross-check. On standard installs it shows Skipped — that's normal, not a failure. |
Stage states
Each stage reads Queued, Running, Done, or Skipped, with progress aggregated over all finalists (validating the top-K means each stage runs once per finalist). Stages run in order; a run is Done when the last stage completes and every finalist has its verdict.
Where you see it
- Runs (sidebar → Runs) — the fleet view: a card per live run with its stages, plus Queued cards showing their queue position, and recent finished runs.
- The run detail — the same stages above the live charts while the run validates.
- Home → Active runs — compact stage summaries across every strategy.
While it validates
Validation is compute-heavy — a big top-K over a long dataset can take a while. You don't have to wait on it: start another strategy's run (extras queue past the scheduler's slot count — see the strategy library), or browse finished results. When the gauntlet finishes, the run's Results stage unlocks and the results table carries a verdict per finalist.
Cancelling during validation keeps completed evaluations — the results table reflects whatever finished, and finalists that never entered the gauntlet simply show no verdict.