Looks like exponents might be an issue

[shamsheersc19]

I tried r^m * ((a - r) / a)^n on the website and got:

Seems strange.

[shamsheersc19]

Another one for x^((a-1)/a)^n :

[pavpanchekha]

Herbie does have this pretty strange habit, yes, of inserting cubes of cube roots. Usually it's overfitting to the sampled input points.