Window 2022-12-02..2026-08-31 | split IS<2025-01-01/OOS>= | OOS baseline +1.11 | SMALL bars: lift>=+0.05, p<0.05, resid>0, IS same sign, skip<50% -> SHADOW sizing only. NOTE: kline cache binds the window (ends 2026-08-31, unchanged vs edge-hunt26) — this run scores stability + SMALL bars, not new power.
- S1 basis-only 0.5/1.25x (lag-1) [FULL]: lift -0.25, p 0.8920, resid -0.49
- S1 basis-only 0.5/1.25x (lag-1) [IS]: lift -0.67, p 0.9268, resid -0.48
- S1 basis-only 0.5/1.25x (lag-1) [OOS]: lift +0.08, p 0.0532, resid +1.18
- S1 unshifted replication [FULL]: lift -0.03, p 0.7756, resid -0.22
- S1 unshifted replication [IS]: lift -0.55, p 0.9076, resid -0.39
- S1 unshifted replication [OOS]: lift +0.38, p 0.0006, resid +1.93
- S2 F&G skip >=75 (lag-1) [FULL]: lift +0.25, p 0.5606, resid +0.14
- S2 F&G skip >=75 (lag-1) [IS]: lift +0.23, p 0.6624, resid +0.26
- S2 F&G skip >=75 (lag-1) [OOS]: lift +0.29, p 0.1484, resid +0.80
- S2: 104 skipped long-days of 364 (28.6%); fg>=75 on 8.8% of all days.
- S1 bucket 0.5x: 7 long-days.
- S1 bucket 1.25x: 14 long-days.
DSR(K=2) S1-vs-baseline OOS: 0.54 | haircut +0.89
S1 basis-only SMALL-track: REJECT (gross lift +0.08 (need >=+0.05); net10 lift +0.06 (need >=+0.05); p 0.0532 (need <0.05); resid +1.18 (need >0); IS lift -0.67 (need same sign); skip 0.0% (need <50%)) S2 F&G skip SMALL-track: REJECT (gross lift +0.29 (need >=+0.05); net10 lift +0.27 (need >=+0.05); p 0.1484 (need <0.05); resid +0.80 (need >0); IS lift +0.23 (need same sign); skip 8.3% (need <50%))
Time 3s. Shadow-only on SHADOW; no live wiring.