Still working on it, it “runs” 22kH but not completely convinced a good portion of that isn’t spitting errors. I’m using a relaxed definition of correct to squeeze higher hashrate out while allowing some rare corner cases to be incorrect. Currently more hashes than expected are incorrect.