That's not really language dependent--that's how floating point numbers always work. And with good reason--there are actually a ton of different NaN bit patterns in the floating point standard. I think anything where the exponent is all 1 is an NaN (except maybe for infinity--I don't remember the exact details). The point is that two completely different patterns of bits can be NaN.

Sure but at least the spec is clear that NaN always returns false.

That is the reason for the existence of the function isNaN.

Plus, NaN returns a generated object, so that result is expected

({}) !== ({})