Ascending / descending triangle (isolated)

Thesis (rejected): Isolating asymmetric triangles from symmetrical pennants — horizontal R with rising S (ascending) or horizontal S with falling R (descending) — should expose the classic "pressure into a wall" geometry that pennants average away.

Code: research/triangle.py Report: research/triangle_report.json (n=1,335)

Edge sign-corrected so positive = pattern direction works.

Note: hit-rate on descending (sign-corrected) is 0.42 — meaning 58 % of descending-triangle breakdowns reverse and go up. That matches the broader "failed breakdown / bear-trap" pattern but is too weak to operationalise here.

Verdict: Reject both subtypes as long/short pattern signals.