Skip to content

Stationarity Validation

Check whether the assumptions that made past data or behavior predictive still hold before extrapolating.

The Diagnostic Story

Symptom: Forecasts, models, baselines, and policies were built from historical data and are now being applied to present decisions as though the process generating that history has not changed. When the results are wrong, the failure is attributed to execution rather than to the assumption that past and present are comparable. Gradually, the gap between the model and reality widens, but because the validation was never explicit, the question of whether it still applies is never clearly raised.

Pivot: Turn assumed stability into an explicit, testable, auditable condition before historical data, models, baselines, or policies are reused for current or future decisions — so continuity with past evidence is preserved where it is warranted, and limited or revised where the process has actually changed.

Resolution: Decisions become better anchored because the usefulness of history is checked rather than assumed. When the process has shifted, the change is visible and the model or baseline can be segmented or updated. The system preserves the value of accumulated evidence without being misled by it.

Reach for this when you hear…

[macroeconomic forecasting] “Our model was trained on the pre-pandemic period and we are still running it without checking whether the relationship between those variables has fundamentally changed.”

[clinical risk scoring] “The scoring tool was validated on a different patient population fifteen years ago — before we extend it to this group we need to check whether the predictive relationships still hold.”

[manufacturing process control] “The control limits are based on baseline data from when this line was running a different material — if the process has drifted, those limits are worse than useless.”

When This Archetype Applies

Complete catalog groundingAt least one sufficient condition set is fully represented by existing primes or domain-specific abstractions.

Past baselines or model behavior guide current decisions even though the process generating observations may have changed.

What this problem means

The structural problem is **unvalidated extrapolation**. A system treats past behavior as predictive because the past was once useful, but the underlying process may no longer be the same. The old baseline can persist through dashboards, operating procedures, contracts, machine-learning models, policy thresholds, or institutional habit. Because the artifact still looks authoritative, users may not notice that the conditions that made it valid have changed.

The key risk is not merely error. It is confidence without comparability: decisions appear evidence-based while quietly relying on a historical regime that no longer describes the present.

Show the applicability expression

Applicability expression2 distinct conditions

Reused prior-valid modelandChanged generating process
Algebraic12

groundedpartly groundedopen

2 conditions, all required.

2Required in every casenumbered 1–2

These hold no matter which pattern applies.

1

Reused prior-valid model · grounded

A model, baseline, forecast, target, alert, report, or policy is reused because it appeared valid under earlier conditions.

2

Changed generating process · grounded

Changes in population, measurement, incentives, tools, environment, behavior, or constraints may have changed the generating process.

Other requirements and context (3)

Why these sit outside the expression

Application gateit governs whether applying the archetype is appropriate or material, rather than defining the structural problem itself.

Supporting contextit may accompany or help interpret the situation, but it is not a load-bearing condition in a sufficient diagnostic set.

  • Application gateA false stability assumption could cause material harm.

  • Supporting contextThe old baseline is embedded in institutional procedures or automation.

  • Supporting contextContext changes faster than revalidation cycles.

2 of 2 conditions grounded.

Read the methodologyDownload the trigger-logic data

Mechanisms / Implementations

  • Stationarity Test: Tests whether the process that generated past lifetimes is still the same process, the precondition for treating survival so far as evidence about survival ahead.
  • Model Drift Monitoring: Watches a live predictor for the slow slide where yesterday's model quietly stops fitting today's world — before the residuals it suppresses start hiding real change.
  • Process Control Chart: Plots a process measurement against statistically derived control limits so ordinary common-cause noise is told apart from the special-cause signals that mean the process has actually shifted off its baseline.
  • Forecast Backtesting: Replays a predictor against withheld history — across time, segments, and regimes — to earn or deny the right to suppress its residuals.
  • Change-Point Detection: Flags the moment the target jumps to a new regime — an abrupt discontinuity the current tracking mode can no longer follow — so the loop switches modes instead of chasing a break as if it were noise.
  • Rolling Window Comparison: Quantifies how much the target, state, and error distributions have drifted by comparing a recent window against earlier ones — turning gradual staleness into a measured magnitude rather than a yes/no event.
  • Baseline Validation Review: A scheduled governance review that decides — before a baseline is reused to set the next round of targets, quotas, or alerts — whether it still describes the world well enough to keep, and records the verdict.
  • Policy Assumption Audit: Re-examines the behavioral and environmental assumptions a standing rule or policy was built on, and narrows or pauses the rule when the world it assumed no longer holds.

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

Built directly on (2)

Also references 14 related abstractions

Variants

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

Model Stationarity Validation · domain variant · recognized

Validates that the assumptions behind a predictive, statistical, or machine-learning model still hold for current use.

Operational Baseline Validation · domain variant · recognized

Checks whether operational baselines, alert thresholds, capacity assumptions, or performance norms remain valid after process or context changes.

Policy Assumption Revalidation · governance variant · recognized

Rechecks whether a policy or rule still rests on valid assumptions about behavior, context, incentives, risks, and institutional capacity.

Measurement Stationarity Check · implementation variant · recognized

Checks whether the measurement process itself remains comparable before interpreting observed changes as real system change.

Regime Shift Gate · temporal variant · candidate

Adds a decision gate when evidence suggests the system has crossed into a materially different operating regime.

Editorial Notes

Problem Classification

Classification: Uncertainty, Evidence & Inference FailureTemporal Process, Nonstationarity & Trend Inference

Problem kernel: past behavior is extrapolated without testing process change

Rationale: Earliest causal condition: A system uses past observations, baselines, model behavior, or policy assumptions to guide present or future decisions even though the process generating those observations may have changed.

Independent corroboration: The earliest necessary condition in the frozen evidence is: A system uses past observations, baselines, model behavior, or policy assumptions to guide present or future decisions even though the process generating those observations may have changed. That is a temporal process nonstationarity and trend inference problem because Historical and sequential evidence is treated as stable, deterministic, or self-explanatory despite drift, dependence, trends, survival conditioning, and time ordering.

Review outcome: Independent reviewer agreement; high confidence.