Integration and reduction

ReduceDimensionality computes weighted or unweighted means and sums over selected axes. axes: non_data reduces leading acquisition dimensions while retaining the trailing scientific rank. Existing uncertainties propagate per name, and optional estimators can add scatter-derived components.

IndexByCoordinate assigns one arbitrary coordinate to one-dimensional bins. It records the physical edges for diagnostics but deliberately does not inspect signals or masks. It replaces the retired scattering-specific IndexPixels step. Existing pipelines must migrate their bin limits and selected coordinate explicitly; see the breaking interface migration guide.

IndexedAverager groups a configured value by that index map. An optional measured axis, such as Q, is averaged over the same accepted points and weights; it is not used to assign bins. The step reports actual mean coordinates, their spread, and population, weight-sum, and effective-sample-size diagnostics rather than substituting nominal bin centres. Bins without positive total weight are omitted and their original IDs are retained in the output. This generic interface replaces the former scattering-specific configuration; in particular, averaging_direction and the use_signal_* fields are no longer accepted.

ConcatenateDatabundles pools compatible one-dimensional bundles before indexed reduction. Its sort_by option is deliberately optional because IndexByCoordinate does not require monotonic input.

FindCenterOfMass1D estimates a one-dimensional peak centre within an iteratively refined contiguous window. It is appropriate when a structured or asymmetric peak should not be forced into a Gaussian model.

Integrate1D performs trapezoidal or Simpson integration of curves on a shared monotonic coordinate axis. It can stably sort a scan and, when explicitly requested, consolidate exact duplicate coordinates before quadrature. Coordinate units multiply into the result unit. Invalid, masked, or zero-weight points define one common domain across all integrated inputs.

Before reducing:

• apply pixel-level corrections that require detector geometry;

• decide whether a mask should mutate the signal or be consumed by the reduction;

• distinguish propagated input uncertainty from scatter estimation; and

• confirm whether weights express statistical importance, exposure, or another reviewed model.