Uncertainty Propagation Model¶
Method — instantiates Monte Carlo Uncertainty Exploration
Propagates uncertainty from input distributions through equations, process logic, or empirical models into output distributions.
Uncertainty Propagation Model owns the transfer function: the explicit map that turns one input vector into one output, and — critically — the structure that says which inputs move together on the way through. Where the sampling engine asks "how do I generate many input vectors and pile up the outputs," this mechanism asks "given a vector, what is the output, and does my map respect that these two inputs are correlated, that this one is bounded, that this nonlinearity bends the average?" Its distinctive claim is that propagation is where dependence and nonlinearity do their damage: feed a model independent draws when the real inputs are coupled, or linearize a curved relationship, and the output distribution will be confidently wrong no matter how many runs you throw at it. It builds and documents the map; it leaves the drawing and aggregating to the engine it rides on.
Example¶
A clinical pharmacology team wants the distribution of peak blood concentration for a new drug across a patient population, to check how often peaks exceed a toxicity ceiling. The propagation model is the pharmacokinetic equation set: dose in, then clearance and volume-of-distribution parameters shape the concentration curve. The subtlety is that clearance and body weight are not independent — heavier patients clear the drug faster — so sampling them separately would invent physiologically impossible patients and understate the high-concentration tail. The model encodes that dependence (a joint distribution over weight and clearance) and carries the nonlinear dose-response curve intact rather than approximating it around the average patient. When many sampled patients are pushed through, the resulting concentration distribution is wider and more skewed on the high side than an independent, linearized version would show — and every one of those choices is written into an assumption record so a reviewer can see exactly which couplings and curvatures were assumed.
How it works¶
- Specify the map explicitly. Encode the equations, process logic, or fitted empirical relationship that turns a sampled input vector into an output, keeping nonlinearities intact rather than linearizing to the mean.
- Model dependence, not just marginals. Represent how inputs co-move — via a joint distribution, a correlation structure, a copula, or hard physical constraints — so impossible combinations are never sampled.
- Pass the mass through. Hand the map to a sampling engine (or, where the algebra allows, an analytic propagation) so the input distributions flow into an output distribution.
- Record the propagation assumptions. Log every dependence assumption, curvature choice, model boundary, and excluded uncertainty, because the output distribution is only as trustworthy as the map that produced it.
Tuning parameters¶
- Model fidelity — a coarse surrogate vs. a high-resolution mechanistic model. Higher fidelity captures more real behavior but is slower to run and harder to inspect.
- Dependence representation — independence, pairwise correlation, full copula, or hard constraints. Richer coupling captures joint risk but demands more evidence to justify.
- Nonlinearity handling — carry the full curve vs. linearize around a point. Linearizing is fast and analytic but silently biases the mean and shrinks the tails.
- Boundary strictness — how aggressively physically impossible combinations are rejected or bounded, trading realism against sampling efficiency.
- Assumption granularity — how finely the record decomposes each modeling choice, trading auditability against overhead.
When it helps, and when it misleads¶
Its strength is that it is the only place where dependence and nonlinearity are handled honestly. Treating coupled inputs as independent, or a curved response as a straight line, is precisely how a model produces a tidy output distribution that is wrong in the direction that hurts — an error the "flaw of averages" names: the output of a nonlinear model at the average input is not the average output.[n1]
Its failure mode is that a rich map invites over-modeling — every added dependency and curve is another assumption that must be defended, and a model too intricate to inspect can hide its own errors behind visual sophistication. The classic misuse is assuming independence for convenience because the joint data is thin, which suppresses exactly the joint tail risk the decision cares about. The guarding discipline is the assumption record: no dependence or curvature enters the map without being written down where a reviewer can challenge it, and any coupling that cannot be evidenced is flagged as an assumption rather than smuggled in as fact.
How it implements the components¶
input_dependency_model— the joint structure, correlations, or constraints that keep sampled input combinations physically and statistically plausible.simulation_model— the explicit equations, process logic, or empirical relationship mapping a sampled vector to an output.assumption_record— the logged dependence choices, curvature decisions, model boundaries, and excluded uncertainties that bound the result's meaning.
It does not run the draw-and-aggregate loop — random_sampling_rule, outcome_distribution, and convergence_diagnostic belong to Monte Carlo Simulation Method. Nor does it calibrate combined risk (calibration_dataset — Portfolio Risk Simulation) or attribute outcome variance across drivers (sensitivity_partition — Stochastic Sensitivity Analysis).
Related¶
- Instantiates: Monte Carlo Uncertainty Exploration — this mechanism supplies the transfer function the archetype's runs flow through.
- Consumes: Monte Carlo Simulation Method — the sampling engine that pushes many drawn vectors through this map.
- Sibling mechanisms: Monte Carlo Simulation Method · Probabilistic Risk Simulation · Scenario Sampling Workflow · Stochastic Sensitivity Analysis · Portfolio Risk Simulation · Operational Capacity Simulation · Simulation Result Dashboard
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Uncertainty Propagation Model operates as an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution because it propagates uncertainty from input distributions through equations, process logic, or empirical models into output distributions.
Independent corroboration: The frozen evidence defines Uncertainty Propagation Model as 'Propagates uncertainty from input distributions through equations, process logic, or empirical models into output distributions', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Universal
Rationale: Uncertainty propagation model is rooted in sampling, inference, measurement, and experimental design; historically, that field developed the core operation described here: propagates uncertainty from input distributions through equations, process logic, or empirical models into output distributions.
Related originating lineages:
- Data Science & Analytics — Data modeling, telemetry, and analytic monitoring supplies a distinct formative lineage for the mechanism's uncertainty propagation model logic.
- Mathematics — Mathematical modeling, proof, and abstract-structure practice supplies a parallel or contributing lineage for the mechanism's defining operation: propagates uncertainty from input distributions through equations, process logic, or empirical models into output distributions.
- Systems Thinking & Cybernetics — Systems thinking, feedback control, and cybernetics supplies a parallel or contributing lineage for the mechanism's defining operation: propagates uncertainty from input distributions through equations, process logic, or empirical models into output distributions.
Review resolution: The blind reviewers agree that statistics_experimental_design is the primary origin and differ only on alternate origin disagreement, origin mode disagreement, domain reach disagreement, encyclopedia synthesis disagreement. I preserve every independently explained alternate from both records rather than imposing a numeric cap. I retain cross_disciplinary_synthesis because the combined evidence shows material contributions from several lineages. The broader reach of universal records portability separately from historical provenance; encyclopedia_synthesis=true preserves the affirmative synthesis judgment where either reviewer identified one.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] A consequence of Jensen's inequality: for a nonlinear function f and a random input X, f(E[X]) ≠ E[f(X)]. Popularized as the "flaw of averages" — plugging average inputs into a nonlinear model gives neither the average nor a representative output, which is why propagation must carry the whole distribution and its dependence structure through the map rather than collapsing to the mean first. ↩