I've got a problem.

Ive got a known unknown, that has a result of a unknown unknown. Its going through a branch predictor of an extremely basic 2-bit saturation.

https://en.wikipedia.org/wiki/Branch...ounter-dia.svg

Its not really working out, the strongly not taken is overwhelming in branch misses, the algo is obviously not predicting properly. Is there a better way without having to rewrite everything, or is it time to get a new AI.