Struct CountsModel 

pub struct CountsModel<'a> {
    pub transmission_model: &'a dyn FitModel,
    pub flux: &'a [f64],
    pub background: &'a [f64],
    pub n_params: usize,
}

Fixed-flux counts-domain forward model: Y_model = flux × T_model(θ) + background.

Retained for the research Fisher helper, not for production fitting. The production counts-KL dispatch (SolverConfig::PoissonKL on InputData::Counts / InputData::CountsWithNuisance) goes through the joint-Poisson conditional-binomial-deviance path in crate::joint_poisson. CountsModel and CountsBackgroundScaleModel below are consumed only by nereids_pipeline::pipeline::evaluate_jacobian_and_fisher (the Fisher-info research helper used by the spatial-regularization epic #394) and by this module’s #[cfg(test)] tests. They assume the caller has pre-computed flux = c · O (i.e. c is baked into flux — a convention that proved error-prone for end users, which is precisely why the production path no longer uses this struct).

Physical count-response contract

This low-level wrapper cannot inspect transmission_model to determine how instrument resolution was applied. Its output must already be the physically valid effective sample/open count response. With response operator R and incident spectrum Φ, the required ratio is

T_eff = R[Φ · T] / R[Φ].

A post-hoc broadened transmission R[T] is not a valid substitute and must never be wrapped here as though it were. In particular, do not directly wrap a resolution-bearing TransmissionFitModel, because that model returns R[T]. An ordinary transmission is valid when resolution is disabled, and a custom inner model that already returns the exact effective ratio is also valid. Otherwise, model the open and sample response arms separately or call the guarded pipeline, which rejects unsupported counts-plus-resolution combinations. When resolution is active, flux must be the matching resolved open-arm response R[Φ].

The flux and background slices must have the same length as the transmission vector returned by the inner model. In debug builds, evaluate() asserts this invariant.

Fields

transmission_model: &'a dyn FitModel

Underlying effective count-ratio model.

See the struct-level physical count-response contract. In particular, this must not be a post-hoc broadened transmission R[T].

flux: &'a [f64]

Open-arm flux response (counts per bin, after normalization).

background: &'a [f64]

Background counts per bin.

n_params: usize

Total parameter count in the wrapped model.

Trait Implementations

impl<'a> FitModel for CountsModel<'a>

fn analytical_jacobian( &self, params: &[f64], free_param_indices: &[usize], y_current: &[f64], ) -> Option<FlatMatrix>

Analytical Jacobian: ∂Y/∂θ = flux · ∂T_inner/∂θ.

Background is constant w.r.t. θ and drops out.

fn evaluate(&self, params: &[f64]) -> Result<Vec<f64>, FittingError>

Evaluate the model for the given parameters.

impl<'a> ForwardModel for CountsModel<'a>

fn predict(&self, params: &[f64]) -> Result<Vec<f64>, FittingError>

Predict model output for the given parameter vector.

fn n_data(&self) -> usize

Number of data points in the model output.

fn n_params(&self) -> usize

Number of parameters (total, including fixed).

fn jacobian( &self, _params: &[f64], _free_param_indices: &[usize], _y_current: &[f64], ) -> Option<Vec<Vec<f64>>>

Analytical Jacobian (column-major layout).