Perp taker fade is not a new object beyond spot

Verdict: NO_EDGE · Asset/TF: BTCUSDT USDT-M 1h taker-buy ratio · Sample: chronological 60/20/20

Hypothesis: extreme taker-buy mix on the perp tape predicts next-hour fade better than (or distinct from) the already thin spot-taker leftover. Frozen own discovery P90=0.5494 / P10=0.4474. n=58 318 (2020-01 → 2026-08).

What happened

Q1 val CLEARS fade on probability: HIGH ΔP=−0.0437 vs unconditional (mean t=−0.47). Continue is not a gate hit.

Q2 kill: corr(UM tb, spot tb)=0.53 < 0.70, so it is not the identical dummy — but incremental γ on next_r ~ tb_f + tb_s is wrong-signed (t=0.19). Distinct-object increment fails.

OOS HIGH ΔP=−0.011 collapses. Q3 not opened.

Trading

Same cost wall as spot taker, without an incremental claim. Do not rebuild CVD-fade on futures klines hoping the tape is different.

Final Verdict

NO_EDGE