Data refresh in 0:10
Experiments
Every arm runs in shadow (simulated fills against live data) so they compare like-for-like; one arm may hold the live paper slot. All P&L is shown gross (fee-free paper) and modelled-net (after Alpaca costs) — a candidate must beat the random-time baseline on net to matter.
Does a $0-commission multi-day equity swing (S8) clear its near-zero cost hurdle net of modelled costs, as the feasibility screen predicts?
| Arm | Trades | Win% (net) | Gross P&L | Cost | Net P&L | Expectancy | Max DD | Markout 30s | vs baseline |
|---|---|---|---|---|---|---|---|---|---|
US Equity/ETF Swing | 0 | 0.0% | $0.0000 | $0.0000 | $0.0000 | $0.00000 | $0.0000 | — | — |
Shadow fills use fair entry/exit prices with all execution cost (fees + spread + slippage) applied via the cost model, so gross mirrors paper and net mirrors live. Results are not like-for-like with real fills — treat as directional evidence, and freeze parameters for 500+ trades before drawing conclusions.
By spread regime
Same comparison, split by the bid/ask spread at entry — an edge that only shows up in tight spreads (or only survives wide ones) is hidden inside the all-time number above.
| Regime | US Equity/ETF Swing |
|---|---|
| tight | — |
| normal | — |
| wide | — |
| unknown | — |
All experiments
| Name | Status | Arms | Live arm | Created |
|---|---|---|---|---|
| default-equity-lab | running | 1 | — | 7/31/2026 |
| default-crypto-lab | running | 10 | — | 7/31/2026 |
Configure arms and the live slot via engine env (CRYPTO_LAB_LIVE_ARM); activate candidate strategies on the strategies page.