Be careful with a doctest example like this:

    >>> foxh([[(1/10,(6,5)), (13/10,1)],[(17/5,2)]],[[(7/5,2)],[(pi,1)]],0.2)
    0.1436702548477872864197888

Here the output is not actually H(1/10, ...) to 25 digits but H(0.100000000000000005551..., ...) to 25 digits.

Here the output is not actually H(1/10, ...) to 25 digits but H(0.100000000000000005551..., ...) to 25 digits.

I guess, you meant "input".

I don't think this issue is too severe nowadays, as Python's float's are usually just IEEE doubles, i.e. doctest will be stable across systems (modulo bugs). But I also suggest using exact binary fractions, not 1/10 or 13/10:
https://mpmath.readthedocs.io/en/latest/basics.html#providing-correct-input

Updated the doc example and tests. The difference in argument indeed causes some deviation in the result.

-    >>> foxh([[(1/10,(6,5)), (13/10,1)],[(17/5,2)]],[[(7/5,2)],[(pi,1)]],0.2)
-    0.1436702548477872864197888
+    >>> foxh([[(mpf('1/10'),(6,5)), (mpf('13/10'),1)],[(mpf('17/5'),2)]],[[(mpf('7/5'),2)],[(pi,1)]],mpf('0.2'))
+    0.1436702548477872392572574

Suggested changes applied.