Rust has precision errors when applying the formula for calculation...

D translation

C++: random tests can generate divisor = 1, which leads to incorrect expected results, and make some valid solutions time out.

Not fixed yet, reraising:

Solutions using floating point numbers may fail because of precision issues, like in
test.assert_equals(divisions(2**100 - 1, 2), 99)
Either there should always be some tests to make sure they always fail or there should not be a non-zero probability of generating such tests to make sure they pass.

And at least in Python the reference solution is incorrect and this case is within the generated range.