There’s a difference between an advance that repudiates prior understanding and one that doesn’t. You can, in maths - and I assume this is the point - know that you are right, in a way that you can’t with a more… epistemological science. Of course it’s more complex than that, and a lot of maths is pretty sciency, like deriving approximate solutions for PDEs is more experimental than you might imagine, but even though we might make improvements there, we’ll never go ‘oh actually those error bounds are wrong’. They might be non optimal but they’ll never be wrong