Uncertainty workflows

Uncertainty modules create or combine named contributions; ordinary BaseData arithmetic propagates them. Read the formal propagation rules before designing a workflow.

PoissonUncertainties estimates counting uncertainty from the signal. Apply it while values still represent counts; after normalization to count rate, propagate the existing component rather than estimating it again.

Keep contributions such as Poisson, readout, transmission, and calibration separate while diagnosing a pipeline. Use CombineUncertainties only for components that the physical model permits combining in quadrature. Use CombineUncertaintiesMax when the intended result is a conservative element-wise envelope.

Reduction modules can estimate uncertainty from observed scatter. That estimate is distinct from propagating known input uncertainty. Configure a collision policy deliberately if both would use the same name.

MoDaCor does not currently represent covariance. Correlated calibration or systematic contributions need an explicit domain model rather than an undocumented quadrature sum.