I'm suggesting guard-banding your specs to account for the imprecision of comparing floating point values. A float64 (double-precision floating-point number) can represent real numbers with approximately 15-17 decimal digits of precision, meaning that while it can represent a very wide range of numbers, there will be a small degree of imprecision in the last few decimal places due to the limitations of binary representation; essentially, most real numbers can only be approximated with a float64, not represented exactly.
This implies the guard-band needs to be at least 1e-15.