Trait LiftParams 

pub trait LiftParams: Real + Sized {
    // Required method
    fn lift(stored: &[f64]) -> Cow<'_, [Self]>;
}

A pricing mode that stored f64 parameters can be lifted into.

Public because Scalar has it as a supertrait, but it is an implementation seam rather than a surface a caller uses: the modes are enumerated below and the set is closed.

Required Methods

fn lift(stored: &[f64]) -> Cow<'_, [Self]>

This mode’s view of stored.

The passive implementation SHALL borrow; every active implementation converts. A caller reads the result as a slice either way.

Dyn Compatibility

This trait is not dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety".

Implementations on Foreign Types

impl LiftParams for f64

fn lift(stored: &[f64]) -> Cow<'_, [Self]>

Implementors

impl LiftParams for AReal<f64>

impl LiftParams for Jet1<f64>

impl LiftParams for Jet2<f64>

impl<const K: usize> LiftParams for JetK<f64, K>

The K-lane forward dual, admitted to the scalar set and consumed by two entry points.

The macro above stamps concrete types; this mode is generic in its lane count, so it is written out. What it does is identical — convert through From<f64>, which gives every lane a zero tangent.

Admission means a generic pricing body can be instantiated at it, and that the cross-mode value-identity harness sweeps it. Consumption is separate, and is a decision per entry point: the K-lane mode serves a full gradient over n inputs in ⌈n/K⌉ tape-free passes, and exactly two entry points present that shape — the calibration Jacobian in calibration.rs and the bond-curve fit in curves/fit.rs, each of which wants every quote against every parameter as a whole matrix in a module that records no tape. Risk stays many-to-one in reverse mode and the pillar solve stays one-to-one in Jet1.

The distinction is a requirement rather than a convention; see the numeric-scalar-modes capability, whose routing test holds the manifest of consumers. What is measured, and what the decision rests on, is benches/mode_crossover.rs.

There is no trait seam for lane seeding. Seedable carries one direction, which is the solve’s shape; each K-lane consumer seeds its own block of K parameters with a unit tangent per lane in a loop it owns, because what is seeded (pillars or coefficients) is the consumer’s business.