Did you read the blog post? They compare to deepswe and call it out as the worst one for false positives (failed, but the benchmark assessed it as correct). It also has less language variance.