Neat approach. I've been dealing with a similar problem in a different domain -- flaky signal detection in trading data. The Bayesian framing is interesting because most trading signal validators just use fixed thresholds. Curious if the entropy-minimization approach would work for narrowing down which candle pattern rules are unreliable vs genuinely noisy.