Window: 2017-11-10 .. 2026-06-26 (2166 daily bars) Cells: 5 (windows=5 * seeds=1) Regime flag: SPY trailing 60d return > 0 -> bull, else bear.
Per-sleeve mean weight (across all cells)
| Sleeve | Agent | EW | delta |
|---|---|---|---|
| eqmom | 0.057 | 0.250 | -0.193 |
| gate60_ETH | 0.042 | 0.250 | -0.208 |
| gate60_BTC | 0.059 | 0.250 | -0.191 |
| spy_bench | 0.010 | 0.250 | -0.240 |
Per-sleeve contribution to mean daily P&L (across all cells)
contrib = mean(w_s * r_s). Sum of all contribs equals mean daily return.
| Sleeve | Agent contrib | EW contrib | delta |
|---|---|---|---|
| eqmom | -0.00001 | +0.00024 | -0.00025 |
| gate60_ETH | +0.00007 | +0.00015 | -0.00007 |
| gate60_BTC | +0.00000 | +0.00013 | -0.00013 |
| spy_bench | +0.00002 | +0.00023 | -0.00022 |
Regime breakdown (mean daily return, aggregate)
| Regime | Agent | EW |
|---|---|---|
| bull | +0.00015 | +0.00076 |
| bear | +0.00003 | +0.00209 |
Correlation structure
- mean(agent - SPY corr): -0.014
- mean(EW - SPY corr): +0.450
- mean(agent - EW corr): +0.069
Interpretation: a positive agent-EW correlation > 0.7 means the agent is mostly a perturbative variant of EW; lower correlation means the agent is making materially different bets. A high agent-SPY correlation means the agent is essentially a levered / de-levered SPY bet.
Per-cell scoreboard
| Window | Seed | OOS start | OOS end | Agent Sharpe | EW Sharpe | agent-SPY corr | agent-EW corr |
|---|---|---|---|---|---|---|---|
| 0 | 0 | 2023-11-15 | 2024-05-16 | +1.17 | +2.44 | -0.031 | -0.001 |
| 1 | 0 | 2024-05-17 | 2024-11-14 | -1.55 | +0.61 | +0.024 | -0.010 |
| 2 | 0 | 2024-11-15 | 2025-05-20 | +2.84 | -0.94 | +0.075 | +0.075 |
| 3 | 0 | 2025-05-21 | 2025-11-18 | +0.84 | +1.81 | -0.097 | +0.102 |
| 4 | 0 | 2025-11-19 | 2026-05-21 | +2.59 | +4.78 | -0.040 | +0.176 |
Verdict (mechanism, not ADOPT/REJECT)
- PASS: agent materially different from EW (agent-EW corr +0.069 < 0.9)
- PASS: agent NOT just a levered SPY bet (agent-SPY corr -0.014, abs < 0.7)
- FAIL: agent does BETTER than EW in bear regimes (agent +0.00003 vs EW +0.00209). This is the hypothesis the plan pre-registered as the regime the agent should win.
elapsed: 101.6s | cells=5