Skip to content

Distributional Assumption Governance

Make probability-distribution commitments explicit, evidence-grounded, consequence-aware, stress-tested, and revisable before they govern inference or action.

Essence

Distributional-Assumption Governance makes commitments about probability shape explicit before they govern inference or action. It treats a proposed family, support, tail, dependence structure, transformation, and observation process as a revisable contract rather than an invisible technical default.

The archetype is needed because a distribution does more than summarize data. It determines which values are possible, how probability moves into unobserved regions, whether extremes are rare or merely unobserved, whether units fail together, and how missing or selected observations are interpreted. Those commitments can change intervals, rankings, reserves, thresholds, safety margins, and who bears error.

Governance does not require one universal test or the most flexible model. It requires evidence proportionate to consequence, credible alternatives, decision-level sensitivity, an enforceable fallback, and a named path for revision when the process or population changes.

Compression statement

When an estimate or decision depends on assuming a particular distributional family, support, tail, dependence, or observation process, define that commitment as a governed contract: specify its use and boundary; justify and diagnose it; compare plausible alternatives; propagate misspecification to the decision; accept, limit, or reject it by stakes-aware criteria; monitor validity; and preserve accountable fallback.

Canonical formula: governed distributional commitment = explicit scope + plausible process + plural diagnostics + alternatives + consequential sensitivity + fit-for-use decision + monitoring + accountable fallback

When to Use This Archetype

Use it when a probability-distribution commitment materially affects an estimate, forecast, interval, simulation, allocation, reserve, threshold, or automated decision. It is especially important when support, zeros, mixtures, tails, dependence, censoring, selection, sparse evidence, or regime change can alter the conclusion.

Use a lighter implementation for reversible, low-stakes description and a stronger implementation for safety, rights, solvency, irreversible investment, or automated decisions. The verification premium should rise when evidence is thin, extrapolation is long, tails dominate harm, or affected groups cannot easily contest an error.

Do not use this archetype merely to decorate a report with diagnostics. If the commitment has no decision consequence, variability characterization may suffice. If evidence cannot justify a strong family, assumption-light inference or bounds may be the appropriate response. If the event class lies outside trusted models, black-swan preparedness remains necessary.

Structural Problem

Many workflows select a familiar distribution before clarifying what quantity, unit, population, horizon, or decision it represents. Software then estimates parameters and reports precision conditional on that choice. The conditional uncertainty looks rigorous while structural uncertainty about family, support, tails, dependence, observation, and regime remains hidden.

Observed shape is not automatically process shape. Aggregation can create smoothness; selection can remove low or high values; censoring can compress tails; mixtures can imitate heavy dispersion; clustering can make independent-looking records share causes. A good fit near the center can coexist with catastrophic error at a decision threshold.

The root tension is productive commitment versus unjustified certainty. Some assumption is often necessary to learn efficiently and act. Yet convenience, precedent, or thin evidence can harden a provisional family into infrastructure. Mature governance preserves useful structure while making its scope, alternatives, consequences, and expiry visible.

Intervention Logic

  1. Lifecycle step 1. Define the decision use, modeled quantity, unit, population, horizon, and consequence of error. The record must show who owns the claim, which evidence supports it, and what downstream action changes if the step fails.
  2. Lifecycle step 2. Record family, support, parameter, tail, dependence, transformation, and observation-process commitments. The record must show who owns the claim, which evidence supports it, and what downstream action changes if the step fails.
  3. Lifecycle step 3. Connect the proposed shape to substantive process and empirical evidence. The record must show who owns the claim, which evidence supports it, and what downstream action changes if the step fails.
  4. Lifecycle step 4. Inspect support, tails, zeros, mixtures, dependence, missingness, censoring, truncation, and selection. The record must show who owns the claim, which evidence supports it, and what downstream action changes if the step fails.
  5. Lifecycle step 5. Compare credible families and assumption-light alternatives. The record must show who owns the claim, which evidence supports it, and what downstream action changes if the step fails.
  6. Lifecycle step 6. Propagate alternatives through decision outputs. The record must show who owns the claim, which evidence supports it, and what downstream action changes if the step fails.
  7. Lifecycle step 7. Validate on simulation, replication, holdout, future, and subgroup evidence where feasible. The record must show who owns the claim, which evidence supports it, and what downstream action changes if the step fails.
  8. Lifecycle step 8. Apply fit-for-use acceptance, exception, or rejection thresholds. The record must show who owns the claim, which evidence supports it, and what downstream action changes if the step fails.
  9. Lifecycle step 9. Activate robust fallback and escalation when evidence is inadequate. The record must show who owns the claim, which evidence supports it, and what downstream action changes if the step fails.
  10. Lifecycle step 10. Monitor distribution shift and retire or revise the commitment. The record must show who owns the claim, which evidence supports it, and what downstream action changes if the step fails.
  11. Lifecycle step 11. Communicate provenance, limitations, affected groups, and downstream uses. The record must show who owns the claim, which evidence supports it, and what downstream action changes if the step fails.

The order matters. A diagnostic cannot establish fitness until the decision use and modeled quantity are stable. An alternative-family comparison cannot protect a decision until differences are propagated into the action. Monitoring cannot preserve validity unless crossing a trigger changes acceptance status and downstream behavior.

Key Components

ComponentDescription
Decision Use and Consequence Scope ↗Defines the inference, forecast, allocation, threshold, or risk decision that will consume the assumed distribution and identifies what changes if the assumption is wrong. Semantic canonical mapping retained the complete legacy component record: {"slug":"decision_use_and_consequence_scope","name":"Decision Use and Consequence Scope","role":"Defines the inference, forecast, allocation, threshold, or risk decision that will consume the assumed distribution and identifies what changes if the assumption is wrong.","notes":"Adequacy is use-relative. A harmless descriptive approximation may be unacceptable when tail probability sets a safety limit, benefit eligibility, capital reserve, or clinical action.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Random Quantity, Estimand, and Unit Definition ↗States exactly what uncertain quantity is modeled, at which unit, aggregation level, time horizon, and population or process boundary. Semantic canonical mapping retained the complete legacy component record: {"slug":"random_quantity_estimand_and_unit_definition","name":"Random Quantity, Estimand, and Unit Definition","role":"States exactly what uncertain quantity is modeled, at which unit, aggregation level, time horizon, and population or process boundary.","notes":"A distribution fitted to transactions, people, sites, days, or averages answers different questions. The unit must not drift between fitting, validation, and decision use.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Data-Generating and Sampling Boundary ↗Documents how observations arise, which cases can enter the data, and which deployment conditions the data are expected to represent. Semantic canonical mapping retained the complete legacy component record: {"slug":"data_generating_and_sampling_boundary","name":"Data-Generating and Sampling Boundary","role":"Documents how observations arise, which cases can enter the data, and which deployment conditions the data are expected to represent.","notes":"Distributional shape can be an artifact of sampling, aggregation, measurement, censoring, selection, or intervention. The boundary separates evidence about observed data from claims about the underlying process.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Distribution-Family Commitment Register ↗Names the proposed family or class, support, discreteness, parameters, tail behavior, dependence claims, transformations, and conditions under which the commitment is asserted. Semantic canonical mapping retained the complete legacy component record: {"slug":"distribution_family_commitment_register","name":"Distribution-Family Commitment Register","role":"Names the proposed family or class, support, discreteness, parameters, tail behavior, dependence claims, transformations, and conditions under which the commitment is asserted.","notes":"The register converts an invisible modeling default into a reviewable contract. It must distinguish fixed structure, estimated quantities, conventions, and unresolved uncertainty.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Empirical Shape and Diagnostic Profile ↗Characterizes center, spread, skew, discreteness, bounds, zeros, modes, outliers, tails, and residual structure using multiple views and uncertainty-aware diagnostics. Semantic canonical mapping retained the complete legacy component record: {"slug":"empirical_shape_and_diagnostic_profile","name":"Empirical Shape and Diagnostic Profile","role":"Characterizes center, spread, skew, discreteness, bounds, zeros, modes, outliers, tails, and residual structure using multiple views and uncertainty-aware diagnostics.","notes":"No single fit statistic is sufficient. The profile should reveal which regions of the distribution drive the intended decision and where data are sparse.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Support, Tail, Zero, and Mixture Review ↗Checks whether the family respects possible values and captures structural zeros, point masses, truncation, heavy tails, multimodality, and latent mixtures. Semantic canonical mapping retained the complete legacy component record: {"slug":"support_tail_zero_and_mixture_review","name":"Support, Tail, Zero, and Mixture Review","role":"Checks whether the family respects possible values and captures structural zeros, point masses, truncation, heavy tails, multimodality, and latent mixtures.","notes":"Good center fit can conceal impossible negative values, understated extremes, or pooled subpopulations. Decision-critical tail and boundary behavior require their own evidence.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Joint Dependence and Conditional Structure ↗Makes independence, conditional independence, clustering, serial dependence, spatial dependence, and cross-variable tail dependence explicit. Semantic canonical mapping retained the complete legacy component record: {"slug":"joint_dependence_and_conditional_structure","name":"Joint Dependence and Conditional Structure","role":"Makes independence, conditional independence, clustering, serial dependence, spatial dependence, and cross-variable tail dependence explicit.","notes":"Correct marginal distributions do not validate a joint model. Dependence errors can dominate pooled risk, uncertainty intervals, and multi-stage decisions.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Missingness, Censoring, Truncation, and Selection Map ↗Traces which values are unobserved, coarsened, excluded, delayed, or preferentially sampled and how that observation process changes the apparent distribution. Semantic canonical mapping retained the complete legacy component record: {"slug":"missingness_censoring_truncation_and_selection_map","name":"Missingness, Censoring, Truncation, and Selection Map","role":"Traces which values are unobserved, coarsened, excluded, delayed, or preferentially sampled and how that observation process changes the apparent distribution.","notes":"Treating observed values as an unbiased sample can turn an observation mechanism into a false shape claim. The map connects data limitations to allowable inference.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Alternative-Family and Assumption-Light Comparator ↗Defines credible rival families, transformations, flexible or nonparametric alternatives, and simple baselines against which the preferred commitment is evaluated. Semantic canonical mapping retained the complete legacy component record: {"slug":"alternative_family_and_assumption_light_comparator","name":"Alternative-Family and Assumption-Light Comparator","role":"Defines credible rival families, transformations, flexible or nonparametric alternatives, and simple baselines against which the preferred commitment is evaluated.","notes":"Governance needs a counterfactual: whether the conclusion depends on this family, any reasonable family, or no strong parametric commitment at all.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Decision-Consequential Sensitivity Map ↗Shows how estimates, rankings, thresholds, intervals, and actions change across plausible distributional commitments and parameter uncertainty. Semantic canonical mapping retained the complete legacy component record: {"slug":"decision_consequential_sensitivity_map","name":"Decision-Consequential Sensitivity Map","role":"Shows how estimates, rankings, thresholds, intervals, and actions change across plausible distributional commitments and parameter uncertainty.","notes":"The key question is not whether every diagnostic rejects, but whether plausible misspecification changes a consequential conclusion. Sensitivity must reach the decision output.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Simulation, Predictive, and Holdout Validation ↗Tests whether the fitted commitment reproduces decision-relevant features in simulated, replicated, held-out, or future observations. Semantic canonical mapping retained the complete legacy component record: {"slug":"simulation_predictive_and_holdout_validation","name":"Simulation, Predictive, and Holdout Validation","role":"Tests whether the fitted commitment reproduces decision-relevant features in simulated, replicated, held-out, or future observations.","notes":"Validation should inspect calibration, residual structure, extremes, subgroup behavior, and operational quantities rather than rewarding in-sample fit alone.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Fit-for-Use Acceptance and Exception Thresholds ↗Defines ex ante evidence thresholds for accepting, conditionally accepting, limiting, escalating, or rejecting a distributional commitment. Semantic canonical mapping retained the complete legacy component record: {"slug":"fit_for_use_acceptance_and_exception_thresholds","name":"Fit-for-Use Acceptance and Exception Thresholds","role":"Defines ex ante evidence thresholds for accepting, conditionally accepting, limiting, escalating, or rejecting a distributional commitment.","notes":"A p-value or information criterion cannot silently decide fitness. Thresholds should reflect stakes, sample size, tail consequence, reversibility, and available alternatives.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Robust Decision Fallback and Escalation Path ↗Specifies safer actions, wider uncertainty, bounds, manual review, assumption-light methods, or deferred commitment when distributional evidence is inadequate. Semantic canonical mapping retained the complete legacy component record: {"slug":"robust_decision_fallback_and_escalation_path","name":"Robust Decision Fallback and Escalation Path","role":"Specifies safer actions, wider uncertainty, bounds, manual review, assumption-light methods, or deferred commitment when distributional evidence is inadequate.","notes":"An audit without an operational fallback merely documents fragility. The response should reduce dependence on unsupported shape claims while preserving necessary decisions.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Regime-Drift Monitoring and Revision Trigger ↗Monitors whether the data-generating process, observation process, population, or dependence structure changes enough to invalidate the accepted commitment. Semantic canonical mapping retained the complete legacy component record: {"slug":"regime_drift_monitoring_and_revision_trigger","name":"Regime-Drift Monitoring and Revision Trigger","role":"Monitors whether the data-generating process, observation process, population, or dependence structure changes enough to invalidate the accepted commitment.","notes":"Distributional adequacy expires. Named triggers, cadence, ownership, and retirement rules prevent a once-reasonable model from becoming an inherited truth.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}
Provenance, Communication, Equity, and Accountability ↗Communicates the assumption, evidence, alternatives, limitations, affected groups, ownership, approvals, and downstream uses in decision-facing language. Semantic canonical mapping retained the complete legacy component record: {"slug":"provenance_communication_equity_and_accountability","name":"Provenance, Communication, Equity, and Accountability","role":"Communicates the assumption, evidence, alternatives, limitations, affected groups, ownership, approvals, and downstream uses in decision-facing language.","notes":"Distributional choices can shift burdens across subgroups, erase rare cases, or manufacture false precision. Traceability and contestability are structural safeguards, not documentation polish.","component_type":"required_structural_component","reuse_scope":"cross_domain","maturity":"provisional","host_archetypes":["distributional_assumption_governance"],"evidence_expectation":"Record the source, uncertainty, owner, decision consequence, review trigger, and evidence needed to retain or revise this part of the assumption contract.","not_a_mechanism_because":"This is a persistent responsibility in governing a distributional commitment, not a particular plot, test, simulation, worksheet, or software procedure."}

Common Mechanisms

MechanismDescription
Distributional-Assumption Card (`distributional_assumption_card`) ↗Type: documentation Summarizes modeled quantity, family, support, parameter treatment, rationale, evidence, limitations, decision use, owner, and revision trigger. Selection constraints and retained implementation evidence: match the data-generating and observation process; reach the decision-critical region rather than center fit only; preserve alternatives, limitations, and revision evidence; Complete legacy mechanism record retained: {"slug":"distributional_assumption_card","name":"Distributional-Assumption Card","mechanism_type":"documentation","role":"Summarizes modeled quantity, family, support, parameter treatment, rationale, evidence, limitations, decision use, owner, and revision trigger.","maturity":"provisional","instantiates_archetypes":["distributional_assumption_governance"],"selection_constraints":["match the data-generating and observation process","reach the decision-critical region rather than center fit only","preserve alternatives, limitations, and revision evidence"],"not_an_archetype_because":"This is a concrete document, diagnostic, comparison, simulation, resampling procedure, stress test, backtest, dashboard, or review gate used within the broader governance lifecycle."}
Support, Shape, and Tail Diagnostic Suite (`support_shape_and_tail_diagnostic_suite`) ↗Type: diagnostic Combines empirical distributions, residual views, quantile comparisons, boundary checks, and tail-focused summaries without treating any one view as decisive. Selection constraints and retained implementation evidence: match the data-generating and observation process; reach the decision-critical region rather than center fit only; preserve alternatives, limitations, and revision evidence; Complete legacy mechanism record retained: {"slug":"support_shape_and_tail_diagnostic_suite","name":"Support, Shape, and Tail Diagnostic Suite","mechanism_type":"diagnostic","role":"Combines empirical distributions, residual views, quantile comparisons, boundary checks, and tail-focused summaries without treating any one view as decisive.","maturity":"provisional","instantiates_archetypes":["distributional_assumption_governance"],"selection_constraints":["match the data-generating and observation process","reach the decision-critical region rather than center fit only","preserve alternatives, limitations, and revision evidence"],"not_an_archetype_because":"This is a concrete document, diagnostic, comparison, simulation, resampling procedure, stress test, backtest, dashboard, or review gate used within the broader governance lifecycle."}
Candidate-Family Comparison Grid (`candidate_family_comparison_grid`) ↗Type: comparison Compares credible families and flexible baselines by support, rationale, predictive adequacy, tail behavior, complexity, and decision consequence. Selection constraints and retained implementation evidence: match the data-generating and observation process; reach the decision-critical region rather than center fit only; preserve alternatives, limitations, and revision evidence; Complete legacy mechanism record retained: {"slug":"candidate_family_comparison_grid","name":"Candidate-Family Comparison Grid","mechanism_type":"comparison","role":"Compares credible families and flexible baselines by support, rationale, predictive adequacy, tail behavior, complexity, and decision consequence.","maturity":"provisional","instantiates_archetypes":["distributional_assumption_governance"],"selection_constraints":["match the data-generating and observation process","reach the decision-critical region rather than center fit only","preserve alternatives, limitations, and revision evidence"],"not_an_archetype_because":"This is a concrete document, diagnostic, comparison, simulation, resampling procedure, stress test, backtest, dashboard, or review gate used within the broader governance lifecycle."}
Predictive Replication Check (`predictive_replication_check`) ↗Type: simulation Generates replicated data from the fitted commitment and compares decision-relevant features with observed or held-out behavior. Selection constraints and retained implementation evidence: match the data-generating and observation process; reach the decision-critical region rather than center fit only; preserve alternatives, limitations, and revision evidence; Complete legacy mechanism record retained: {"slug":"predictive_replication_check","name":"Predictive Replication Check","mechanism_type":"simulation","role":"Generates replicated data from the fitted commitment and compares decision-relevant features with observed or held-out behavior.","maturity":"provisional","instantiates_archetypes":["distributional_assumption_governance"],"selection_constraints":["match the data-generating and observation process","reach the decision-critical region rather than center fit only","preserve alternatives, limitations, and revision evidence"],"not_an_archetype_because":"This is a concrete document, diagnostic, comparison, simulation, resampling procedure, stress test, backtest, dashboard, or review gate used within the broader governance lifecycle."}
Resampling Robustness Audit (`resampling_robustness_audit`) ↗Type: resampling Re-estimates conclusions under resampling or perturbation to reveal finite-sample instability and dependence on a few observations. Selection constraints and retained implementation evidence: match the data-generating and observation process; reach the decision-critical region rather than center fit only; preserve alternatives, limitations, and revision evidence; Complete legacy mechanism record retained: {"slug":"resampling_robustness_audit","name":"Resampling Robustness Audit","mechanism_type":"resampling","role":"Re-estimates conclusions under resampling or perturbation to reveal finite-sample instability and dependence on a few observations.","maturity":"provisional","instantiates_archetypes":["distributional_assumption_governance"],"selection_constraints":["match the data-generating and observation process","reach the decision-critical region rather than center fit only","preserve alternatives, limitations, and revision evidence"],"not_an_archetype_because":"This is a concrete document, diagnostic, comparison, simulation, resampling procedure, stress test, backtest, dashboard, or review gate used within the broader governance lifecycle."}
Distributional Sensitivity Grid (`distributional_sensitivity_grid`) ↗Type: sensitivity_review Propagates plausible family, tail, dependence, transformation, and parameter choices through the final decision output. Selection constraints and retained implementation evidence: match the data-generating and observation process; reach the decision-critical region rather than center fit only; preserve alternatives, limitations, and revision evidence; Complete legacy mechanism record retained: {"slug":"distributional_sensitivity_grid","name":"Distributional Sensitivity Grid","mechanism_type":"sensitivity_review","role":"Propagates plausible family, tail, dependence, transformation, and parameter choices through the final decision output.","maturity":"provisional","instantiates_archetypes":["distributional_assumption_governance"],"selection_constraints":["match the data-generating and observation process","reach the decision-critical region rather than center fit only","preserve alternatives, limitations, and revision evidence"],"not_an_archetype_because":"This is a concrete document, diagnostic, comparison, simulation, resampling procedure, stress test, backtest, dashboard, or review gate used within the broader governance lifecycle."}
Tail and Boundary Stress Scenario (`tail_and_boundary_stress_scenario`) ↗Type: stress_test Tests actions against heavier tails, structural zeros, truncation, mixtures, extremes, and support-boundary behavior that ordinary fit can hide. Selection constraints and retained implementation evidence: match the data-generating and observation process; reach the decision-critical region rather than center fit only; preserve alternatives, limitations, and revision evidence; Complete legacy mechanism record retained: {"slug":"tail_and_boundary_stress_scenario","name":"Tail and Boundary Stress Scenario","mechanism_type":"stress_test","role":"Tests actions against heavier tails, structural zeros, truncation, mixtures, extremes, and support-boundary behavior that ordinary fit can hide.","maturity":"provisional","instantiates_archetypes":["distributional_assumption_governance"],"selection_constraints":["match the data-generating and observation process","reach the decision-critical region rather than center fit only","preserve alternatives, limitations, and revision evidence"],"not_an_archetype_because":"This is a concrete document, diagnostic, comparison, simulation, resampling procedure, stress test, backtest, dashboard, or review gate used within the broader governance lifecycle."}
Holdout Calibration and Coverage Backtest (`holdout_calibration_and_coverage_backtest`) ↗Type: validation Checks predictions, intervals, ranks, or event rates on data not used to choose or fit the distributional commitment. Selection constraints and retained implementation evidence: match the data-generating and observation process; reach the decision-critical region rather than center fit only; preserve alternatives, limitations, and revision evidence; Complete legacy mechanism record retained: {"slug":"holdout_calibration_and_coverage_backtest","name":"Holdout Calibration and Coverage Backtest","mechanism_type":"validation","role":"Checks predictions, intervals, ranks, or event rates on data not used to choose or fit the distributional commitment.","maturity":"provisional","instantiates_archetypes":["distributional_assumption_governance"],"selection_constraints":["match the data-generating and observation process","reach the decision-critical region rather than center fit only","preserve alternatives, limitations, and revision evidence"],"not_an_archetype_because":"This is a concrete document, diagnostic, comparison, simulation, resampling procedure, stress test, backtest, dashboard, or review gate used within the broader governance lifecycle."}
Distribution-Shift Trigger Dashboard (`distribution_shift_trigger_dashboard`) ↗Type: monitoring Tracks shape, residual, tail, subgroup, missingness, and dependence indicators against revision thresholds. Selection constraints and retained implementation evidence: match the data-generating and observation process; reach the decision-critical region rather than center fit only; preserve alternatives, limitations, and revision evidence; Complete legacy mechanism record retained: {"slug":"distribution_shift_trigger_dashboard","name":"Distribution-Shift Trigger Dashboard","mechanism_type":"monitoring","role":"Tracks shape, residual, tail, subgroup, missingness, and dependence indicators against revision thresholds.","maturity":"provisional","instantiates_archetypes":["distributional_assumption_governance"],"selection_constraints":["match the data-generating and observation process","reach the decision-critical region rather than center fit only","preserve alternatives, limitations, and revision evidence"],"not_an_archetype_because":"This is a concrete document, diagnostic, comparison, simulation, resampling procedure, stress test, backtest, dashboard, or review gate used within the broader governance lifecycle."}
Independent Assumption-Challenge Gate (`independent_assumption_challenge_gate`) ↗Type: governance Requires an independent reviewer to inspect rationale, alternatives, consequential sensitivity, safeguards, and exception decisions before high-stakes use. Selection constraints and retained implementation evidence: match the data-generating and observation process; reach the decision-critical region rather than center fit only; preserve alternatives, limitations, and revision evidence; Complete legacy mechanism record retained: {"slug":"independent_assumption_challenge_gate","name":"Independent Assumption-Challenge Gate","mechanism_type":"governance","role":"Requires an independent reviewer to inspect rationale, alternatives, consequential sensitivity, safeguards, and exception decisions before high-stakes use.","maturity":"provisional","instantiates_archetypes":["distributional_assumption_governance"],"selection_constraints":["match the data-generating and observation process","reach the decision-critical region rather than center fit only","preserve alternatives, limitations, and revision evidence"],"not_an_archetype_because":"This is a concrete document, diagnostic, comparison, simulation, resampling procedure, stress test, backtest, dashboard, or review gate used within the broader governance lifecycle."}
  • Candidate-Family Comparison Grid
  • Distribution-Shift Trigger Dashboard
  • Distributional Sensitivity Grid
  • Distributional-Assumption Card
  • Holdout Calibration and Coverage Backtest
  • Independent Assumption-Challenge Gate
  • Predictive Replication Check
  • Resampling Robustness Audit
  • Support, Shape, and Tail Diagnostic Suite
  • Tail and Boundary Stress Scenario

Parameter / Tuning Dimensions

  • Decision stakes: Consequence, reversibility, rights, safety, and scale determine evidence and review depth.
  • Modeled unit: Individual, group, site, interval, transaction, or aggregate changes the implied process.
  • Family rigidity: Named parametric family, flexible class, mixture, semi-parametric, or nonparametric comparator.
  • Support contract: Bounded, nonnegative, discrete, continuous, circular, zero-inflated, censored, or truncated values.
  • Tail emphasis: How much validation effort reaches rare but consequential regions.
  • Dependence scope: Serial, spatial, hierarchical, network, conditional, and tail dependence included.
  • Alternative breadth: Number and plausibility of rival families and assumption-light baselines.
  • Extrapolation distance: How far the decision reaches beyond observed values, populations, or regimes.
  • Acceptance threshold: Evidence needed for accepted, conditional, limited, escalated, or rejected status.
  • Sensitivity tolerance: How much output or decision movement is acceptable across credible commitments.
  • Fallback conservatism: Bounds, wider intervals, robust action, delay, manual review, or reserve.
  • Validation separation: Degree of protection between selection, fitting, tuning, and evaluation data.
  • Monitoring cadence: Frequency and power of shift, tail, residual, subgroup, and missingness review.
  • Revision latency: Time allowed between trigger crossing and downstream model change.
  • Communication depth: Technical record, decision memo, public limitation notice, or affected-party explanation.
  • Independent challenge: Who may contest evidence, exceptions, and burden distribution.

Tune dimensions jointly. A rigid family may aid interpretation but magnify misspecification; a flexible model may fit better but obscure mechanism and extrapolation. Stronger monitoring can detect drift earlier but create false alarms. The right balance is governed by the cost of error, not by methodological fashion.

Invariants to Preserve

  • The modeled quantity, unit, boundary, and decision use remain stable and traceable.
  • Support and impossible values are respected.
  • Tail and dependence claims do not outrun evidence.
  • Credible alternatives remain visible.
  • Diagnostics reach decision consequences.
  • Exceptions produce enforceable fallback.
  • Subgroup and rare-case harms remain reviewable.
  • Acceptance has ownership, expiry, and revision triggers.
  • Conditional estimates must never be communicated as unconditional certainty.
  • Failure to reject a diagnostic null must never be treated as proof of the assumed family.
  • No accepted commitment may lose its owner, provenance, exception status, or expiry trigger when embedded downstream.

These invariants protect both epistemic and operational integrity. They allow the specific family, mechanism, and threshold to change while preserving inspectability, contestability, and safe response to misspecification.

Target Outcomes

  • Fewer hidden probability-model defaults.
  • More defensible uncertainty and extrapolation.
  • Decisions robust to plausible misspecification.
  • Earlier detection of tail, dependence, selection, and regime failures.
  • Clearer model limitations and fallback actions.
  • Improved reproducibility and accountability.
  • Less false precision and rare-case erasure.

Success is not a claim that the true distribution has been discovered. It is evidence that the chosen commitment is explicit, proportionate to its use, compared with credible alternatives, validated where consequences concentrate, and linked to action when evidence weakens.

Tradeoffs

  • Model parsimony versus distributional flexibility: resolve the balance using decision consequence, data limits, explanatory plausibility, downstream dependence, reversibility, and the distribution of error burdens.
  • Interpretability versus predictive adequacy: resolve the balance using decision consequence, data limits, explanatory plausibility, downstream dependence, reversibility, and the distribution of error burdens.
  • Stronger assumptions versus information efficiency: resolve the balance using decision consequence, data limits, explanatory plausibility, downstream dependence, reversibility, and the distribution of error burdens.
  • Center fit versus tail fidelity: resolve the balance using decision consequence, data limits, explanatory plausibility, downstream dependence, reversibility, and the distribution of error burdens.
  • Pooled stability versus subgroup validity: resolve the balance using decision consequence, data limits, explanatory plausibility, downstream dependence, reversibility, and the distribution of error burdens.
  • Analytic convenience versus support realism: resolve the balance using decision consequence, data limits, explanatory plausibility, downstream dependence, reversibility, and the distribution of error burdens.
  • Early decision versus additional validation: resolve the balance using decision consequence, data limits, explanatory plausibility, downstream dependence, reversibility, and the distribution of error burdens.
  • Narrow precision versus misspecification robustness: resolve the balance using decision consequence, data limits, explanatory plausibility, downstream dependence, reversibility, and the distribution of error burdens.
  • Standardization versus domain-specific rationale: resolve the balance using decision consequence, data limits, explanatory plausibility, downstream dependence, reversibility, and the distribution of error burdens.
  • Monitoring sensitivity versus false alarms: resolve the balance using decision consequence, data limits, explanatory plausibility, downstream dependence, reversibility, and the distribution of error burdens.
  • Open disclosure versus security or privacy: resolve the balance using decision consequence, data limits, explanatory plausibility, downstream dependence, reversibility, and the distribution of error burdens.
  • Automation speed versus independent challenge: resolve the balance using decision consequence, data limits, explanatory plausibility, downstream dependence, reversibility, and the distribution of error burdens.

The governance process itself has cost. Additional alternatives, validation, and independent review consume time and expertise. But efficiency should be assessed against the cost of hidden misspecification, rework, unsafe thresholds, unfair exclusions, and inherited models whose limitations can no longer be traced.

Failure Modes

Software Default Commitment

Cause. A familiar package or default family substitutes for substantive rationale and alternatives.

Mitigation. Require an explicit family register, mechanism link, and independent comparison. Confirm that remediation changes acceptance status or downstream action rather than only adding documentation.

Single Metric Fit Absolutism

Cause. One test, score, or plot is treated as universal proof of adequacy.

Mitigation. Use plural diagnostics and fit-for-use thresholds tied to decision-critical features. Confirm that remediation changes acceptance status or downstream action rather than only adding documentation.

Center Fit Tail Failure

Cause. Good average fit hides understated extremes, structural zeros, bounds, or mixtures.

Mitigation. Run tail, support, zero, and mixture review plus stress scenarios. Confirm that remediation changes acceptance status or downstream action rather than only adding documentation.

Marginal Fit Joint Failure

Cause. Marginal distributions look adequate while dependence or clustering invalidates joint conclusions.

Mitigation. Model and validate conditional, serial, spatial, clustered, and tail dependence explicitly. Confirm that remediation changes acceptance status or downstream action rather than only adding documentation.

Observation Process Blindness

Cause. Missingness, censoring, truncation, selection, or measurement creates the apparent shape.

Mitigation. Map the observation process and limit inference to supported boundaries. Confirm that remediation changes acceptance status or downstream action rather than only adding documentation.

Diagnostic Data Reuse

Cause. The same data select, fit, tune, and validate the family, producing optimistic adequacy.

Mitigation. Use holdout, replication, resampling, or protected validation evidence where feasible. Confirm that remediation changes acceptance status or downstream action rather than only adding documentation.

Decision Disconnect

Cause. Diagnostics are reported but plausible alternatives are never propagated to the actual decision.

Mitigation. Build a consequential sensitivity map and define action thresholds. Confirm that remediation changes acceptance status or downstream action rather than only adding documentation.

False Precision Under Weak Evidence

Cause. Sparse data are converted into narrow estimates by an unjustified family commitment.

Mitigation. Widen uncertainty, use bounds or assumption-light comparators, and escalate irreversible uses. Confirm that remediation changes acceptance status or downstream action rather than only adding documentation.

Subgroup And Rare Case Erasure

Cause. A pooled distribution fits the majority while harming or misrepresenting smaller groups and rare cases.

Mitigation. Validate by affected subgroup, preserve tail cases, and document burden shifts. Confirm that remediation changes acceptance status or downstream action rather than only adding documentation.

Stationary Forever Assumption

Cause. A once-accepted family persists after the process, population, measurement, or dependence structure changes.

Mitigation. Monitor regime evidence and use named revision and retirement triggers. Confirm that remediation changes acceptance status or downstream action rather than only adding documentation.

Assumption Documentation Theater

Cause. A model card exists, but unsupported commitments still drive automated or irreversible decisions.

Mitigation. Link exception status to enforceable fallback, approval, and downstream-use controls. Confirm that remediation changes acceptance status or downstream action rather than only adding documentation.

Method Over Archetype Drift

Cause. The intervention collapses into a catalog of tests or plots without ownership and decision governance.

Mitigation. Preserve the lifecycle: scope, commitment, evidence, alternatives, consequence, acceptance, monitoring, and accountability. Confirm that remediation changes acceptance status or downstream action rather than only adding documentation.

Neighbor Distinctions

coverage_probability_calibration

Calibrates whether an interval procedure achieves its nominal coverage. Distributional-Assumption Governance is broader and upstream: it governs family, support, tail, dependence, observation-process, alternative, and retirement commitments across many outputs.

knowledge_warrant_audit

Audits the type and strength of warrant for beliefs generally. This archetype specializes that discipline into the structure and consequences of probability-distribution commitments.

assumption_light_inference

Selects methods that require fewer fragile assumptions. It is one fallback or comparator; the parent also governs when a stronger distributional commitment is justified and how it remains accountable.

assumption_stress_testing

Breaks or reverses plan assumptions to test resilience. This archetype examines a specific statistical commitment through fit, alternatives, decision sensitivity, validation, and monitoring.

uncertainty_explicitness

Makes unknowns, ranges, and confidence visible. Distributional governance determines whether the probability structure used to produce those representations is defensible.

variability_characterization

Describes observed variation before acting. This archetype governs inferential commitments about the data-generating and observation processes, including extrapolation and decision use.

stationarity_validation

Checks whether predictive conditions remain stable through time or regime. Stationarity is one distributional premise; the parent also covers support, family, tails, mixtures, dependence, and selection.

sensitivity_analysis_protocol

Varies assumptions or parameters to reveal conclusion fragility. Distributional sensitivity is a mechanism inside a broader contract that also includes rationale, acceptance, fallback, provenance, and retirement.

generalization_validation

Tests whether a learned pattern transfers to new cases. Distributional governance may use holdout evidence but also addresses whether the assumed probability form itself is plausible and decision-fit.

representative_sampling_design

Designs observation selection to represent a target population. Sampling is an input boundary here; even representative samples can be modeled with an unsuitable family or dependence structure.

black_swan_preparedness

Builds survival capacity for high-impact surprise outside trusted forecasts. Distributional governance improves known model commitments but must not claim to eliminate deep uncertainty or unknown event classes.

tail_risk_preservation

Protects rare important cases from common-case optimization. Tail-case protection is a safeguard here, but the parent governs the full distributional model lifecycle, including ordinary, joint, and limited-data settings.

The decisive boundary is lifecycle ownership. Related archetypes may supply uncertainty representation, assumption reduction, stress tests, coverage calibration, warrant review, sampling, stationarity, or generalization evidence. Distributional-Assumption Governance integrates these selectively around one specific commitment and its decision consequences.

Cross-Domain Examples

Clinical Trials

Govern a response-distribution and dropout commitment before using a treatment-effect interval for approval.

Support, missingness, subgroup, tail, and decision thresholds materially affect safety and benefit claims. The implementation records its modeled unit, family and observation assumptions, alternatives, decision sensitivity, acceptance status, fallback, and revision trigger.

Insurance And Finance

Govern severity, frequency, and dependence assumptions used to size reserves and concentration limits.

Tail and joint misspecification can produce solvency decisions far outside observed samples. The implementation records its modeled unit, family and observation assumptions, alternatives, decision sensitivity, acceptance status, fallback, and revision trigger.

Supply And Service Operations

Govern demand and lead-time distributions used for inventory and staffing thresholds.

Zeros, bursts, seasonality, truncation, and regime shifts change shortage and excess-capacity tradeoffs. The implementation records its modeled unit, family and observation assumptions, alternatives, decision sensitivity, acceptance status, fallback, and revision trigger.

Machine Learning

Govern residual, class, and deployment-shift assumptions used to produce predictive uncertainty.

Held-out center performance can conceal subgroup, tail, dependence, and shift failures. The implementation records its modeled unit, family and observation assumptions, alternatives, decision sensitivity, acceptance status, fallback, and revision trigger.

Environmental Risk

Govern precipitation or loss-tail extrapolation used for infrastructure design levels.

Sparse extremes, nonstationarity, threshold choice, and irreversible consequence demand explicit limits and stress bounds. The implementation records its modeled unit, family and observation assumptions, alternatives, decision sensitivity, acceptance status, fallback, and revision trigger.

Public Policy

Govern a benefit or burden distribution used to set eligibility, enforcement, or resource thresholds.

Pooled fit can hide subgroup burden and turn modeling convention into distributive policy. The implementation records its modeled unit, family and observation assumptions, alternatives, decision sensitivity, acceptance status, fallback, and revision trigger.

Extended emergency-service example

A regional emergency service uses a probability model to set staffing and supply capacity. The team first defines calls per zone-hour as the modeled quantity and distinguishes routine demand from declared incidents. It records count support, excess zeros, seasonality, spatial clustering, response-time censoring, and population changes. Several count, mixture, flexible, and heavy-tail alternatives are justified and compared. The team examines predictive replication, holdout coverage, extreme-day performance, neighborhood slices, and dependence during shared events. Each credible alternative is propagated through missed-service and overtime decisions. Where the action changes, managers adopt a robust staffing floor and escalation reserve rather than choosing the most convenient family. Acceptance is conditional, with independent review, drift triggers, quarterly revalidation, and a public limitation note. The distribution is therefore useful without being treated as timeless truth.

Across domains, the mathematical family changes but the intervention remains stable: expose the commitment, test what matters for use, compare alternatives, propagate uncertainty into action, and revise when evidence or conditions change.

Non-Examples

  • Selecting a normal distribution because a software dialog preselects it
  • Running one normality test and declaring the entire model valid
  • Plotting a histogram without linking shape to inference or action
  • Using a nonparametric test without governing the decision boundary
  • Publishing a model card whose exception status has no operational effect
  • Claiming a tail model predicts every black swan

A fit result is not governance, a caveat is not a fallback, and an assumption is not validated forever. The archetype exists only when evidence, decision consequence, acceptance, accountability, and revision are connected.

Abstractions this archetype builds on — directly (a source ingredient) or as a related pattern. Links follow the typed catalog namespace.

Built directly on (1)

Also references 17 related abstractions

Variants

Narrower or domain-specific specializations that share this archetype's core structure. Recognized variants are established; candidate variants are provisional.

Parametric-Family Selection Governance · mechanism family variant · recognized

Govern selection among named parametric families when interpretable parameters, compact models, or analytic decisions make a family commitment useful.

  • Distinct from parent: The parent also governs flexible, semi-parametric, tail, mixture, and dependence commitments.
  • Use when: a named family supports the intended inference; sample size and mechanism provide discriminating evidence; family choice materially affects estimates or decisions.
  • Typical domains: experimental science, forecasting, operations, econometrics
  • Common mechanisms: distributional assumption card, candidate family comparison grid, predictive replication check

Tail and Extreme-Distribution Governance · risk or failure variant · recognized

Govern commitments about rare, extreme, bounded, or heavy-tailed outcomes when tail shape drives safety, reserve, or loss decisions.

  • Distinct from parent: Many distributional decisions concern center, discreteness, zeros, or ordinary predictive fit rather than extremes.
  • Use when: extremes dominate consequence; ordinary samples contain little tail evidence; threshold, reserve, or safety decisions depend on extrapolation.
  • Typical domains: insurance, infrastructure safety, climate risk, cybersecurity
  • Common mechanisms: support shape and tail diagnostic suite, tail and boundary stress scenario, distributional sensitivity grid

Joint-Dependence Distribution Governance · scale variant · recognized

Govern joint-distribution and dependence assumptions when correlated observations, clustered systems, or conditional relationships drive conclusions.

  • Distinct from parent: Univariate settings may not require a full joint-dependence contract.
  • Use when: marginal fit can conceal joint failure; observations cluster across time, place, group, or network; pooling or portfolio claims depend on independence.
  • Typical domains: portfolio risk, public health, supply networks, spatial analysis
  • Common mechanisms: predictive replication check, distributional sensitivity grid, holdout calibration and coverage backtest

Limited-Data Distribution Governance · risk or failure variant · recognized

Govern distributional commitments when small samples, censoring, missing tails, shifting regimes, or weak observability prevent decisive family validation.

  • Distinct from parent: Better-observed settings may support firmer fit-for-use acceptance.
  • Use when: evidence cannot distinguish plausible families; important regions are missing or censored; decisions must proceed despite irreducible shape uncertainty.
  • Typical domains: rare disease, new products, disaster planning, emerging technology
  • Common mechanisms: resampling robustness audit, tail and boundary stress scenario, independent assumption challenge gate

Near names: Distributional Assumption Audit, Shape-Family Commitment Review, Parametric Assumption Review.