I don’t think he did? Psychohistory relies on gargantuan groups of people acting predictably, so you can’t predict a one-off random mutation of a single individual. That’s not a failure, it’s a limitation- and why whenever Seldon appeared, he prefaced his statements with a % chance that the plan was still succeeding- the remaining failure case included a catastrophic mule-like event.

Mule-like occurrences were always a possibility, which is why Seldon also created the Second Foundation to fix the bug and set the plan back on track- which of course they did. Things were a bit touch-and-go for a while there, but the possibility of failure is always a consequence of probabilistic modeling

Not a faliure, but a predictable weakness of the model that was well known and was brought up several times. Psychohistory cannot model individual eccentricities, the mule was a huge divide-by-zero trap for the equations. Seldon had the forsight to write an exception handler in the form of the Second Foundation.