changed that test. i ran the full tests a bunch of time with the reference solution and the user solution using different order of operations, and tests passed consistently, so these problematic cases should be exceedingly rare.

Still got the same issue here in cpp

I think it also makes a difference if you calculate as w/(h*h) or w/h/h

The test cases should not sit on floating point boundary.

Almost a year later, still the same issue.

And actually, the test is not even an edge-case which causes weird floating point rounding in typical implementations. The test-case is actually simply wrong:

w = 86.7000000000000028421709430404007434844970703125
h = 1.6999999999999999555910790149937383830547332763671875

so w>86.7, h<1.7, so w/(h* h) > 86.7/(1.7^2) = 30. So Obese is actually correct, given the input.