Right, but that's what makes me doubt this statement in the comment I was replying to

> The probability that SynthID changes the best choice to a worse one is equal to the probability that it changes a worse choice to the best one.