Pre-registered 2026-09-09. Ports Study-5 graduates (E3 band, E1 blend) EXACTLY to the live eqmom ranking (resid126 sector-demeaned) on the live 101-name universe (99 matched), top_k=10, band=14 (1.4x top_k — scaled port of 70/50, zero re-tuning). Next-open, 10bps one-way, sqrt(252). Split 2016-01-01. Look-ahead guard PASS.
| Variant | OOS Ret | OOS Sharpe | OOS MaxDD | Turn/yr | CorrMkt | CorrBaseLive | ResidSh | p vs BASE_live |
|---|---|---|---|---|---|---|---|---|
| BASE_raw101 | +1110.71% | 1.11 | -34.75% | 8.66 | 0.75 | 0.97 | 0.48 | ref |
| BASE_live resid+sector H21 | +1507.27% | 1.29 | -30.87% | 8.24 | 0.75 | 1.00 | 0.74 | - |
| E1p H-blend live | +1504.64% | 1.29 | -31.65% | 6.35 | 0.77 | 0.99 | 0.73 | 0.5012 |
| E3p hyst-14 live | +1207.51% | 1.19 | -30.86% | 6.32 | 0.76 | 0.99 | 0.60 | 0.9536 |
IS (pre-2016) reference
| Variant | IS Ret | IS Sharpe | IS MaxDD |
|---|---|---|---|
| BASE_raw101 | +468.11% | 0.90 | -46.82% |
| BASE_live resid+sector H21 | +284.38% | 0.68 | -62.35% |
| E1p H-blend live | +290.54% | 0.69 | -62.92% |
| E3p hyst-14 live | +310.20% | 0.70 | -62.88% |
Verdict rules (pre-registered): PORT PASS = OOS p<0.05 vs BASE_live (Bonferroni ~=0.025 for 2 tests noted) with turnover cut >=15% (or DD cut >=5pp), resid sane, IS consistent. Then recommend live adoption. Else REJECT (no transfer; live stays).
Verdicts (2026-09-09)
- E3p hyst-14 — REJECT (no transfer). OOS 1.19 vs BASE_live 1.29, p=0.95; resid 0.60 < 0.74. Turnover cut (-23%) is real but buys a -0.10 Sharpe drag. Mechanism lesson: hysteresis value is inversely related to ranking quality. On noisy raw ranking (Study 5) it blocked bad swaps (+0.05); on the sharp resid+sector ranking it blocks GOOD swaps — holding incumbents decayed to rank 11-14 dilutes a selector that already picks better. The device does not transfer. Live stays unchanged. Study-5 E3 remains valid only for raw-ranked books (none live).
- E1p H-blend — PORT PASS (efficiency fallback). OOS 1.29 vs 1.29, p=0.50 (exact tie), turnover 6.35 vs 8.24 (-23%), resid 0.73, IS consistent (0.69 vs 0.68). 25bps stress: 1.240 vs 1.224 (holds). Same shelf as H42/H63: approved fallback if costs rise, not a replacement.
- Reference read: BASE_live 1.29 vs BASE_raw101 1.11 OOS confirms the live resid+sector promotion on the long 2016+ window (+0.18). Honest caveat: the ranking inverts IS (0.68 vs raw 0.90) — its edge concentrates post-2016, consistent with book_expand R5 (edge concentrates post-2019, 4/5 windows). Not an action item, but the live ranking's pre-2016 softness is now on record.
Ledger: no live-code change. Live eqmom (resid+sector H21, 10 names) stands; E1p joins the fallback shelf. Next genuinely-new selection ideas must beat 1.29 on this 101/live-ranking harness — turnover devices are mined out.