Yes, precisely downwards nearest multiple.
The original idea is whether a function can return a value to a given precision (something like floor, round).
Furthermore, it will be more useful if it works for more general precision, say 0.05, 0.02, 0.03...