A three-asset crowded-long book is weaker, not stronger

Verdict: WATCH · Book: nonoverlap HIGH shorts on BTC+ETH+SOL · Hold: 8h · Sample: chronological 60/20/20

Hypothesis after the BTC nonoverlap demotion: pooling more names restores |t|≥2 with more independent bets. Frozen parent P90s (BTC 2.572, ETH 2.965, SOL 4.009), 8h cooldown per asset.

What happened

Q1 FAIL. Pooled val: n=730, mean +11.8 bp, t=1.22. Mean still > 0 → not NO_EDGE. Adding names hurt power vs BTC-only leftover (n=71, +43.5 bp, t=1.60).

Why: ETH under the frozen discovery P90 is a crowded-long regime (HIGH rate 39.6%, mean run 53h). Event pooling is ETH-weighted (425/730) with dependent streak samples. Date-clustered t=0.91. Per-asset val t: BTC 1.60 / ETH 0.50 / SOL 0.66.

Q2: ETH+SOL alone +8.4 bp, t=0.81 — not BTC-only, still underpowered. OOS +21.8 bp, t=1.65 — sign holds, not OVERFIT. Q3 val 5 bps t=0.35; 10 bps mean flips negative.

Trading

Do not “fix” the single-position demotion by adding ETH/SOL under discovery-frozen P90s. You buy more dependent ETH hours, not more independent edge.

Final Verdict

WATCH

Same crowded-long mean residue. Diversification across these three tails does not create a book.