Double bottom / double top

Thesis (rejected): Two pivot lows (or highs) within 15-80 trading days, prices within 2 ATR of each other, with a mid-rally / mid-dip prominence of ≥ 2 ATR, that then break the neckline on volume, reverse direction for 5-40 days.

Code: research/double_bottom_top.py Report: research/double_bottom_top_report.json (n=31,116)

Why it failed: Edges are basically zero in both directions.

Even the deduplicated detector (each p1 produces at most one event, each (pattern, p2) produces at most one event) shows no edge. Pivot- to-pivot retests of the same level are common enough — and the volume-confirmed neckline break is loose enough — that the cohort captures a sample of mostly random breakouts.

Verdict: Skip. Could potentially be salvaged by tightening to "first significant" pivot pairs only (e.g. through a ZigZag indicator with a 5-10 % move filter) but the unconditional signal is dead.