The verified falsification record — reproducible trading strategies, out-of-sample stress tests, and negative findings published with complete code, seeds, and evidence.
Crowded-long ∩ 15m vol filter: val n=38 mean +66 bp but t=1.41 — louder, not stronger than parent t=1.60. WATCH.
SergBTC+ETH+SOL nonoverlap HIGH shorts: val n=730 mean +11.8 bp t=1.22 — weaker than BTC-only. ETH regime dilutes. WATCH.
SergBTC sum_toptrader L/S: val HIGH−MID t=−0.95; corr vs global 0.16 (distinct); HIGH rate drifts 10%→68%. NO_EDGE.
SergBTC HIGH-only nonoverlap: n=71 val t=1.60 fails |t|≥2; mean still ~44 bp; OOS sign holds. Demote overlapping PROMOTED to WATCH.
SergBTC top-trader L/S HIGH: val fade t=−2.17 but corr vs global 0.99; incremental γ t=−1.45. Same costume. NO_EDGE.
SergBTC HIGH-only crowded-long short: val +41.5 bp at 5 bps (t=3.93), OOS sign holds; not redundant with OI. Clustered overlap caveat. PROMOTED research object — not live.
SergETH/SOL global L/S HIGH: val t=−2.06 / −3.13, not redundant with |r|; ETH OOS holds; Q3 ETH ≤0 at 0 bps. WATCH, not a book.
SergBTC global L/S ratio: crowded-long underperforms val t=−5.14 / OOS t=−2.16, not redundant with OI/funding; shorts do not bounce. WATCH, not PROMOTED.
Serg