When would knowing that they're an approximation ever be inadequate compared to knowing why they're an approximation in the real world for the simple masses? 9

When would knowing that they're an approximation ever be inadequate compared to knowing why they're an approximation in the real world for the simple masses? 99.99% of cases the decision goes "floats are fast and memory efficient, but I need full decimal precision, so I'll use the library"