Function hankel_usable_from 

pub fn hankel_usable_from<E, V, const FLOOR_N: LargeInt, const FLOOR_D: LargeInt>(
    nu: V,
) -> V
where
    E: FloatElement,
    V: FloatVector<Element = E>,

The smallest x at which the Hankel arm reaches full precision for order nu.

Only ever used as a gate, so being slightly conservative is free and being optimistic is not. nu was swept to 12 when this was fitted. Past that the uniform Debye regime takes over and this form should not be trusted. See the table in the module documentation.

FLOOR_N / FLOOR_D is the format’s constant floor, which comes from $e^{-2x} < \varepsilon$ and is therefore per-format: <17, 1> for binary64 and <13, 2> for binary32, both measured rather than derived. It is a const generic for the same reason N is: the caller knows its format and the value should be materialized, not computed.