BTC open gaps vanished and never picked a side

Verdict: NO_EDGE · Asset/TF: BTCUSDT 1d (primary) and 4h · Sample: chronological 60/20/20

Hypothesis: overnight/session gap (open/prev_close − 1) either fades into the bar or continues through it more than chance.

What happened

Large gap = |gap| ≥ discovery P90 (1d thr ≈ 0.8 bps, 4h ≈ 0.7 bps). On validation/OOS the 1d series has zero large events; 4h has one. Gaps collapse toward ~0 after ~2023 on chained Binance klines — the feature almost disappears.

Where events still exist (discovery): continue rate ≈ 0.48–0.52 vs 0.5; sign-shuffle null; signed expected bar return ≈ 0. Non-prereg per-split decile sensitivity tells the same story.

Trading

Q3 skipped. No edge to cost-test.

Final Verdict

NO_EDGE

There is no stable fade/continue premium in BTC open gaps on this sample, and the modern regime barely has gaps to trade. Do not build gap strategies from 24/7 crypto klines without a different gap definition (e.g. weekend halt, exchange-specific).