Research Lab Findings
Every strategy the research engine has tested. Ordered by validation P&L on held-out data. We publish the losers too, not just the winners. That's the whole point.
How the research lab works
- Candidate strategies get queued automatically (entries across RSI, MFI, StochRSI, Williams %R, CCI, ROC, BB, Donchian).
- Each candidate runs against 3 years of Binance data on 10 coins at 5m timeframe with real fees.
- Walk-forward validation splits data 60% train / 40% test. In-sample numbers that don't survive out-of-sample are flagged.
- Strategies pass through one of three approval lanes: (a) standard: positive walk-forward P&L, 15+ trades, 30%+ WR, Deflated Sharpe fluke probability < 50%. (b) near-breakeven with strong fundamentals: P&L > -0.02% with 100+ trades, 60%+ WR, fluke < 50%. (c) high-conviction selective: 8+ trades with 75%+ WR and clear positive P&L. We tightened these lanes on 2026-04-30 after a gate audit; 549 prior approvals were retroactively retired as noise.
- Whatever survives graduates to forward testing. Whatever fails is published here with the fail reason.
See the full engine architecture →Is the regime classifier even working? →
About what you see below
Losing strategies show every detail. We publish the specific indicators, thresholds, and parameters that didn't work because the honest map of what failed is more useful than most research results. Passing strategies show the indicator family and a coarse performance bucket. The exact thresholds, timeframes, and exit params for winners stay private while we forward-test them ourselves. Edges die when they get crowded. When one graduates to real-capital deployment we publish the full post-mortem with performance data that's already baked in.
What passed (1 patterns)
Exact params redacted until deployed. Patterns grouped by indicator family.
sma_21
15m · long
| Entry | Exit | Val P&L | Val WR | Val trades | Sharpe | Fluke % | Confidence | Status |
|---|---|---|---|---|---|---|---|---|
| sma_21 crosses above sma_50 | fee-baked 2.5% net TP, 627-bar time, ADX<50 | +2.198% | 100% | 31 | 12.51 | 0% | thin | holdout failed |
| sma_21 crosses above sma_50 | fee-baked 2.5% net TP, 655-bar time, ADX<50 | +2.198% | 100% | 31 | 11.56 | 0% | thin | holdout failed |
| sma_21 crosses above sma_50 | fee-baked 2.5% net TP, 686-bar time, ADX<50 | +2.198% | 100% | 31 | 11.71 | 0% | thin | holdout failed |
| sma_21 crosses above sma_50 | fee-baked 2.5% net TP, 540-bar time, ADX<50 | +2.198% | 100% | 31 | 11.51 | 0% | thin | holdout failed |
| sma_21 crosses above sma_50 | fee-baked 2.5% net TP, 509-bar time, ADX<50 | +2.198% | 100% | 31 | 12.01 | 0% | thin | holdout failed |
| sma_21 crosses above sma_50 | fee-baked 2.5% net TP, 607-bar time, ADX<50 | +2.198% | 100% | 31 | 12.64 | 0% | thin | holdout failed |
See a combination that interests you? Test your own variant with real fees:
Test a strategy →Research findings are from our ongoing automated backtesting. Walk-forward validation splits data 60%/40%. “Passed” means all 5 statistical gates cleared. Numbers shown are on held-out (40%) validation data. Not financial advice.