Wyckoff spring / liquidity-sweep reclaim

Thesis (rejected): Inside an established 30-day trading range (height ≤ 20 %), a day that pierces the range low intraday (wick ≥ 0.5 % below range_low) but closes back inside the range on volume ≥ 1.2× — the "spring" — should outperform on the bounce hypothesis. Distinct from failed_breakdown.py (multi-day reclaim).

Code: research/wyckoff_spring.py Report: research/wyckoff_spring_report.json (n=24,124)

n is large and every cohort is flat-to-negative. Even the deepest drawdown bucket (which lifts every other contrarian pattern in this file) is only flat on this one.

Why it failed: The same-day reclaim intrabar mechanic is real in microstructure, but our data is daily close — by the time we see the reclaim, the bounce has happened and the residual forward window is exactly what controls capture. failed_breakdown works because it requires a separate reclaim day after the breakdown closes through support, which is a different (and more selective) event.

Verdict: Reject. The "spring" effect is intraday; daily-close detection captures the wrong slice of it.