The gauntlet
Every strategy in Forven moves through four stages. Nothing trades until it has earned its way to the end.
Stages
Screen. Cheap synthetic checks. Does the strategy produce any signal at all? Does it survive a coarse backtest on a universe sample? Most strategies fail here — which is the point.
Gauntlet. Walk-forward out-of-sample testing across multiple regimes, with fresh windows the strategy has never touched. Pass thresholds are set per asset, not global.
Candidate. The strategy runs paper live for a minimum window. Slippage, latency, and fill assumptions get re-tested against real book state.
Live. Capital allocated via Hyperliquid routing. Post-mortems run automatically on every closed trade.
Graduation ratios
A typical cohort looks like:
stage in out kept
────────────────────────────────────
screen 1000 280 28%
gauntlet 280 42 15%
candidate 42 9 21%
live 9 2 22%
────────────────────────────────────
graduation ratio: 2 / 1000 (0.2%)
This is illustrative. Your ratios depend on universe, regime, and parameter discipline. The point: most strategies fail — and Forven kills them before you do.
Why walk-forward
In-sample metrics are useless for evaluating strategies you plan to trade. Walk-forward out-of-sample is the only honest proxy for forward performance that doesn't require waiting months of real time to learn you were wrong.
Forven records IS and OOS separately. You'll see both in the lab. Trust OOS.