Same universe/split/costs/fill/liq-gate as expand2-4. Judged by testbot_baseline.judge.
| Variant | IS Sharpe | IS maxDD | OOS Sharpe | OOS maxDD | OOS ret | turn/yr |
|---|---|---|---|---|---|---|
| U1 PEAD-momentum blend | 0.85 | -54.8% | 1.75 | -22.7% | +148.6% | 10.8 |
judge: U1 PEAD-momentum blend vs frozen baseline v1 (overlap 5030d)
| leg | base Sharpe | cand Sharpe | base maxDD | cand maxDD | | IS | 0.88 | 0.85 | -54.7% | -54.8% | | OOS | 1.74 | 1.75 | -28.5% | -22.7% | IS bootstrap p=0.825 | IS p<0.05: NO | OOS Sharpe >= base: OK | OOS DD within +5pp: OK verdict: FAIL — move on
| U2 TLT proxy | 0.88 | -42.2% | 1.19 | -28.9% | +112.7% | 8.8 |
judge: U2 TLT proxy vs frozen baseline v1 (overlap 5030d)
| leg | base Sharpe | cand Sharpe | base maxDD | cand maxDD | | IS | 0.88 | 0.88 | -54.7% | -42.2% | | OOS | 1.74 | 1.19 | -28.5% | -28.9% | IS bootstrap p=0.656 | IS p<0.05: NO | OOS Sharpe >= base: NO | OOS DD within +5pp: OK verdict: FAIL — move on
Addendum: static BASE/TLT blends (post-hoc, same gate)
| Variant | IS Sharpe | IS maxDD | OOS Sharpe | OOS maxDD | verdict |
|---|---|---|---|---|---|
| U3 80/20 | 0.99 | -43.5% | 1.70 | -23.1% | FAIL (IS p=0.93, OOS -0.04) |
| U4 60/40 | 1.12 | -30.7% | 1.58 | -17.7% | FAIL (IS p=0.93, OOS -0.16) |
Verdicts (2026-09-09)
ADOPT: nothing. Scoreboard: 26 judged variants, 0 adopted.
- U1 PEAD blend — FAIL. IS tie (p=0.825), OOS Sharpe tied with better DD but lower return. Combining the two surviving edges at the ranking level adds nothing here. (Sleeve-level combination — separate PEAD sleeve next to the momentum book — is a different, untested construction.)
- U2 TLT proxy — FAIL cleanly. Bonds in low-breadth spells cut IS DD but collapse OOS Sharpe (1.19 vs 1.74): 2023-26 punished duration while equities ripped. Timing into TLT is rejected; static blends below are the residual question.
- U3/U4 static blends — FAIL on the bar, best insurance prints yet. U4: IS Sharpe +0.24 with DD nearly halved. But the gate's IS leg tests mean(diff)>0, and blends win via the volatility denominator, not the mean (p=0.93) — plus OOS Sharpe pays (1.58 vs 1.74). Disclosed limitation of the gate, not a bug: diversifiers are judged on DD-cut-per-Sharpe-cost by operator judgment, never auto-PASS. U4 joins vol-target/credit/breadth in the insurance family: protection that costs in equity bull windows.