Function discount_bond_affine 

pub fn discount_bond_affine<R: Scalar>(
    params: &HullWhiteParameters<R>,
    curve: &dyn YieldCurve<R>,
    t: NaiveDate,
    maturity: NaiveDate,
) -> Result<(R, R), CurveError>

The affine factors of the model’s discount bond seen from t: P(t, T; r) = A(t, T) · e^{−B(t, T)·r}, with B = (1 − e^{−a(T−t)}) / a and A = P(0,T)/P(0,t) · exp(B·f(0,t) − σ²(1 − e^{−2at}) B² / 4a).

Everything is in the mode — the curve ratio, the forward, the parameters — so a bond price evaluated at a lattice node carries the curve’s and the parameters’ sensitivities. Freezing the forward passively, as an earlier lattice asset did, dropped the ∂P/∂curve term through it at every exercise node.

Errors

Whatever the curve reports for t, T or the forward stencil.