Function spatial_map_typed 

pub fn spatial_map_typed(
    input: &InputData3D<'_>,
    config: &UnifiedFitConfig,
    dead_pixels: Option<&Array2<bool>>,
    cancel: Option<&AtomicBool>,
    progress: Option<&AtomicUsize>,
) -> Result<SpatialResult, PipelineError>

Fit every pixel of a 3-D data cube and return per-pixel maps — the spatial-mapping entry point of the pipeline.

Runs the single-spectrum fitter once per (y, x) pixel of input (shape (n_energies, height, width)), in parallel over pixels with rayon, and assembles the results into SpatialResult: one areal density map and uncertainty map per fitted isotope/group, the χ² (or deviance-per-dof) map, the convergence mask, and any optional maps the configuration enables (temperature, normalization, background terms, t0 / flight-path scale).

Input modes

input selects the per-pixel objective: pre-normalized InputData3D::Transmission (+ per-bin uncertainty), InputData3D::Counts (sample + open-beam), or InputData3D::CountsWithNuisance (sample + flux + background nuisance arms; counts-domain solvers only).

Validation (all up-front, before any pixel is fitted)

• The cube’s spectral axis must match config.energies(), the mode’s companion cubes must match the primary cube’s shape, and dead_pixels (when given) must be (height, width).
• Cube values are validated on live pixels, each against its domain — transmission finite, uncertainty finite and strictly positive, counts/flux finite and non-negative, background finite — so a corrupt cube fails loudly instead of producing a quietly-NaN map. For transmission inputs with a fit_energy_range, the value checks are scoped to the active bins — out-of-range bins may contain NaN by design.
• Known-degenerate configurations are rejected with a diagnostic rather than letting every pixel fail into an all-NaN map: counts + LM + fit_energy_scale (numerically ill-conditioned per-pixel — issue #458 B3) and CountsWithNuisance with an LM solver (requires a counts-domain solver). transmission_background settings are validated here for the same reason. (fit_energy_scale together with fit_temperature is SUPPORTED since issue #634 — the energy-scale model carries a fitted temperature column.)

Per-pixel fit failures after validation are not errors: the pixel is recorded as NaN in the maps, converged_map is false there, and n_failed counts it.

Cancellation and progress

cancel is polled before the sweep and at every pixel; once set, remaining pixels are skipped and the call returns PipelineError::Cancelled (partial results are discarded). progress is incremented once per completed live pixel, so a UI thread can poll it against the number of live pixels (height × width minus the dead_pixels-masked count).

Errors

PipelineError::ShapeMismatch for axis/shape disagreements, PipelineError::InvalidParameter for rejected configurations and invalid cube values, PipelineError::Transmission when the shared cross-section / resolution-plan precompute fails (e.g. a resolution-kernel or working-grid build error), and PipelineError::Cancelled when cancel was set.