Actually if you use Decimal("...") instead of float("...") for your calculations, just converting the final result back to float type works.

I agree, not cool to test sizes explicitely indicated as not being tested ... =)
I took me time to see what the problem was, especially as we have no access to which input values are tested.