propagate-uncertainty
Purpose
Propagate input uncertainty through a model or reasoning chain to estimate output distributions, tails, and contributions.
Input contract
required: [model, uncertain_inputs, output_variables]
optional: [sampling_plan, dependency_structure, tail_metrics]
constraints: [input distributions, dependencies, and sampling method must be declared]
Procedure
- Represent each uncertain input with its distribution and dependencies.
- Sample or propagate the inputs through the model.
- Summarize output distributions, tail behavior, and dominant uncertainty contributors.
Output contract
produces: [output_distributions, tail_summary, uncertainty_contributions]
delta_fields: [findings, evidence_updates, uncertainties]
Quality gates
- Output uncertainty is traceable to input distributions and sampling assumptions; tails are not replaced by point estimates.
Failure and counterexamples
Do not claim propagated uncertainty when dependencies or model validity are unknown.
Provenance map
deep-insight/uncertainty-cascade: resolved.monte-carlo-propagation: concept (no exact pool entry).
