gpd-error-propagation

Error propagation traces uncertainty from inputs through intermediate results to final quantities by computing sensitivities, handling correlations, and identifying dominant sources to generate a complete error budget.

An error budget is a structured record of all uncertainty sources—input, numerical, approximations, and model—needed when multi-step derivations risk overconfident results. It is required for auditability in scientific workflows and computational modeling.

Uncertainty propagation through a derivation chain handles correlations by computing sensitivities analytically or numerically, tracing combined input uncertainties through each intermediate calculation step to the final output quantity.

Error propagation applies directly across phase-based derivations in scientific workflows, tracing inputs to final quantities and updating project state records with propagated uncertainties for full auditability.

The best way to generate an error budget is running a propagation workflow that identifies all uncertainty sources, computes sensitivities, handles correlations, and outputs an ERROR-BUDGET.md alongside updated state records.

Sensitivity analysis misses numerical and approximation uncertainties when it only evaluates input variations. Complete error propagation identifies all sources—input, numerical, approximations, and model—computing sensitivities to produce a full budget.