Failed-pattern fade survey

Thesis (mixed result): failed_breakdown (failed bearish breakdowns through support) is a strong long signal. The symmetric test: do failed bullish breakouts and failed bearish-pattern neckline breaks produce edges in the opposite direction?

Source events are the existing report event lists for coil_breakout, head_shoulders, and double_bottom_top. For each event we check whether the price closes back through the pattern's pivot within 5 days; if so, that day is the fade entry. Direction is signed so that positive edge = fade thesis works.

Code: research/failed_pattern_fade.py Report: research/failed_pattern_fade_report.json (n=18,046 total)

Reading: the edge column is sign-multiplied by direction, so positive = the fade thesis works. coil_failed is the only cohort with a meaningful, monotone edge (~ +1.0 %/+1.7 %/+1.2 % over 10/20/40 days going short).

The other three cohorts are flat to marginal: hs_top_failed is actually mildly bearish (the failed top breakdown keeps reverting back down, not continuing up). double_top_failed and double_bot_failed are within 0.4 % of zero.

Verdict: Keep coil_failed as a candidate SHORT signal — it is the mirror of failed_breakdown and adds the first SHORT-side pattern alpha to the repo other than bearish_divergence. Reject the H&S and double-top/bottom mirrors.