PRIMARY = skip iff z24>+0.5 (live ETH setting, no re-tuning), pooled equal-weight across parity-passing alts. No house IS/OOS split (alt depth starts 2025-07-15 — IS<2025 would be empty); verdict on full window + 4-cut walk-forward. Costs gross + net10/25 per unit-notional flip. Decide code-path parity by construction (frozen gate60.decide, eth slot); study bars resample Binance klines while live reads Alpaca (ETH-study caveat).
Window: 2022-12-01 .. 2026-08-31 (1370 days, 405 traded, 64 blocked). Coins pooled: SOL, XRP, LINK, AVAX, DOGE, ADA.
Parity gate (a): z24 distribution vs ETH backtest (mean0, std1)
| Coin | z24 mean | z24 std | Parity |
|---|---|---|---|
| SOL | +0.011 | 1.069 | PASS |
| XRP | +0.007 | 1.093 | PASS |
| LINK | +0.042 | 1.109 | PASS |
| AVAX | +0.048 | 1.163 | PASS |
| DOGE | +0.027 | 1.085 | PASS |
| ADA | +0.020 | 1.109 | PASS |
Pooled scoreboard (PRIMARY)
- gross lift +0.34, bootstrap p 0.603
- beta vs BH: -0.11 | corr: -0.45 | resid Sharpe +0.16 (ret +5.01%, maxDD -31.68%)
- blocked 4.7% | turnover 729 flips
- DSR(K=6) 0.98 | haircut +1.04
- net10 Sharpe +1.27 (base +1.02)
Verdict: REJECT (lift +0.34 >= +0.10; p 0.603 NOT < 0.05; resid +0.16; blocked 4.7%; DSR(K=6) 0.98; WF 0/4 (need 3))
Walk-forward (4 cuts, frozen rule, lift>=+0.10 + p<0.05)
| Cut | OOS days | Traded | Base Sh | Gated Sh | Lift | p | Pass |
|---|---|---|---|---|---|---|---|
| 2025-11-01 | 78 | 31 | -1.06 | -0.78 | +0.27 | 0.342 | N |
| 2026-02-01 | 92 | 0 | +nan | +nan | +nan | nan | N (VOID: no exposure — all-alt flat regime) |
| 2026-05-01 | 89 | 0 | +nan | +nan | +nan | nan | N (VOID: no exposure — all-alt flat regime) |
| 2026-08-01 | 92 | 0 | +nan | +nan | +nan | nan | N (VOID: no exposure — all-alt flat regime) |
Forward returns of pooled OK long days by z24 quartile
| Quartile (z24, pooled longs) | Mean daily ret | Days |
|---|---|---|
| Q1(low) | +0.951% | 459 |
| Q2 | +0.541% | 459 |
| Q3 | +0.500% | 458 |
| Q4(high) | -0.061% | 459 |
IS/OOS diagnostic (house split; information only)
| Segment | Days | Traded | Base Sh | Gated Sh | Lift | p |
|---|---|---|---|---|---|---|
| IS (<2025-01-01) | 762 | 282 | +1.62 | +1.88 | +0.26 | 0.812 |
| OOS (>=2025-01-01) | 608 | 123 | +0.08 | +0.55 | +0.46 | 0.229 |
Reading
- PASS needs all bars incl. WF 3/4. VOID cuts (no exposure) count as N under the frozen rule — the WF component here is uninformative (one tradeable cut: lift +0.27, p=0.34), so the REJECT rests on the full-window p=0.603 alone.
- DATA NOTE: alt depth caches self-extended during this run (
load_depth_bars(days=0)fetched 2023-01+ from Binance vision S3; caches now 2023-01-01..2026-09-12, fresh). No lookahead: day-boundary sampling + lag-1 alignment unchanged; pre-depth days remain WARMING/blocked. Side benefit: the live alt-depth feed problem is smaller than assessed (still needs a refresher loop — no--loopexists yet). - REGIME NOTE: gate60 held ZERO alt exposure Nov 2025..Aug 2026 (9-month all-alt flat — crypto bear, BTC ~$77k Aug 2026). The pooled lift lives entirely in the pre-Nov-2025 window.
- No live wiring (sym_map + live_runner_alt hook) and no Binance refresher on this result; alt-depth parked (OKX alt recording is the only other future feed).
Time: 60s. Cap: skip-only filter, no sizing.