Taker-buy extremes fade a little, then fees eat it

Verdict: NOT_TRADEABLE · Asset/TF: BTCUSDT 1h (primary), 15m (secondary) · Sample: chronological 60/20/20

Hypothesis: extreme taker-buy imbalance (aggressor mix, not raw volume size) predicts the next bar — crowding fade or continuation. Distinct from the prior volume-spike ablation.

What happened

Existing OHLCV had no taker columns → Binance spot klines from data.binance.vision with taker-buy fields.

Q1 (1h val): fade clears — HIGH ΔP −6.6 pp, LOW +4.3 pp vs unconditional.

Q2: probability fade persists inside sign(r_t) strata (not just “up bars have high taker-buy”); OLS on validation is weak.

OOS 1h: same-sign fade, attenuated (ΔP_HIGH −2.4 pp) — still holds directionally. 15m corroborates.

Trading

Q3 fade book ≈ 1.6 bps gross (val) / 1.0 bps (OOS). Negative at 5–10 bps.

Final Verdict

NOT_TRADEABLE

Fourth member of the thin-fade family (close-location, streak, Donchian, taker-buy). Real enough to clear validation; too thin for retail round-trip costs. Do not ship a CVD-extreme fade bot on BTC klines at 5 bps.