I don't understand this. You don't check a math proof against evidence, you check it for the correctness of the logical deductions from the axioms to the conclusions.
Here's a proof that there are infinitely many primes:
Take any finite collection of primes. Multiply them all together, and add one, giving a number X. Let P be the smallest prime that divides X (note that P might equal X). P cannot be in the original collection, so we have created a new prime. Therefore any finite collection of primes is incomplete, and so the collection of all primes must be infinite.
Suppose I want to check that for correctness. No amount of evidence will show that it's correct, so I really don't know what you are saying. Perhaps you could expand on your thoughts using this as a specific example.
If human-performed math were not a science, a retraction would never happen.