| Cell | Fade | EW | Uplift | maxDD | turn |
|---|---|---|---|---|---|
| v6_main_w0_s0 | -0.812 | -0.820 | +0.008 | -17.04% | 0.0002 |
| v6_main_w0_s1 | -0.816 | -0.820 | +0.004 | -17.11% | 0.0005 |
| v6_main_w0_s2 | -0.817 | -0.820 | +0.003 | -16.91% | 0.0003 |
| v6_main_w0_s3 | -0.807 | -0.820 | +0.013 | -17.27% | 0.0002 |
| v6_main_w0_s4 | -0.813 | -0.820 | +0.007 | -17.10% | 0.0002 |
| v6_main_w1_s0 | +1.793 | +1.810 | -0.017 | -8.84% | 0.0004 |
| v6_main_w1_s1 | +1.815 | +1.810 | +0.004 | -8.81% | 0.0000 |
| v6_main_w1_s2 | +1.815 | +1.810 | +0.005 | -8.66% | 0.0002 |
| v6_main_w1_s3 | +1.800 | +1.810 | -0.010 | -8.94% | 0.0002 |
| v6_main_w1_s4 | +1.802 | +1.810 | -0.008 | -9.13% | 0.0002 |
| v6_main_w2_s0 | +4.941 | +4.974 | -0.032 | -3.10% | 0.0003 |
| v6_main_w2_s1 | +4.982 | +4.974 | +0.008 | -3.03% | 0.0002 |
| v6_main_w2_s2 | +4.976 | +4.974 | +0.002 | -3.11% | 0.0000 |
| v6_main_w2_s3 | +4.984 | +4.974 | +0.011 | -3.09% | 0.0002 |
| v6_main_w2_s4 | +4.970 | +4.974 | -0.004 | -3.24% | 0.0000 |
median uplift +0.004 | pooled p=0.329
- uplift>0.10: NO
- p<0.05: NO
- net alpha: OK
- maxDD>-30%: OK
- turnover<=10%: OK
verdict: REJECT
Interpretation (2026-09-09)
The fade is the mirror image the theory predicts: agent −0.053, fade +0.004, both bootstrap-indistinguishable from zero (p=0.495 / 0.329). The +0.004 even has the expected sign — fading recovers roughly the agent's own turnover drag while paying a mirror drag. There is no systematic wrongness to exploit in either direction; the tilts are variance, and variance has no opposite trade. Fading only works on BIASED errors (significant negative edge net of the fade's own costs), and five studies agree none exists here. The momentum-short leg (T7, IS −0.67) already showed what "trade against momentum" costs in this program. Case closed on both sides of EW.