I fail to see your point. Floats are best practice for many financial applications, where model error already eclipses floating point error and performance matters.

Your point in the floating-point discussion is true, but you’re wrong about this one- linguistics is a descriptive field, not a prescriptive one.