The Protocol

The Global Curiosity Engine (GCE) is a falsification-first protocol for locating where long-horizon predictive
certainty breaks. It does not explain causes, produce forecasts, or recommend actions. It identifies
Predictive Stress Zones (PSZ)—places where a predictive assumption (typically smooth trend continuity)
fails to remain stable under controlled perturbations—and returns only disciplined, assumption-focused questions.

Core idea

  • Detection finds candidate stress signals using simple, transparent metrics.
  • Falsification tries to break those signals using perturbations (windows, thresholds, revisions/datasets).
  • Only method-invariant stress survives. Anything brittle is discarded or downgraded.
  • Output is inquiry, not interpretation. The protocol produces questions that target assumptions.

Inputs

  • Domain dataset: a public time series suitable for rolling baselines (e.g., climate normals, fertility rates).
  • Geographic or categorical index: states, countries, regions, divisions, cohorts, etc.
  • Variable(s): the measured quantity (e.g., temperature, precipitation, TFR).
  • Time base: annual, monthly, or fixed-interval series (prefer consistent cadence).

Primary stress signal (v1.0): Trend instability

In GCE v1.0, the primary stress signal is directional trend reversal across rolling baselines.
For each location and variable, the protocol estimates the direction of the trend (increase, decrease, or stall)
within a baseline window. A candidate PSZ is flagged when the direction reverses between successive
baselines (e.g., decrease → increase, increase → decrease).

Operational definitions

  • Baseline window: a rolling window of fixed length (e.g., 20 or 30 years) used to estimate local trend direction.
  • Directional trend: sign of the slope (or equivalent) over the window, with a defined stall threshold.
  • Reversal: a sign change (or direction change) between adjacent windows.
  • Predictive Stress Zone (PSZ): a location/variable/time-span where reversals persist after falsification.

Falsification gates (non-negotiable)

A candidate PSZ is not accepted until it survives the following perturbations. The protocol prefers
conservative thresholds; the goal is to discard brittle findings.

Gate 1 — Window perturbation

  • Re-run detection using multiple baseline lengths (e.g., 20-year and 30-year windows).
  • Stress test whether reversals persist or vanish when window size changes.

Gate 2 — Threshold sensitivity

  • Vary the stall threshold and reversal sensitivity (e.g., small vs. stricter slope thresholds).
  • Confirm that major reversals persist; minor reversals that appear/disappear are documented as sensitivity artifacts.

Gate 3 — Cross-revision / cross-dataset consistency

  • When available, compare against an alternate revision vintage (e.g., WPP 2022 vs WPP 2024) or a second dataset.
  • If a reversal appears only in one revision/dataset and not the other, it is downgraded or discarded.

Procedure (step-by-step)

  1. Ingest the dataset and normalize it into a long-form table: location × time × variable.
  2. Construct rolling baseline windows across the available timeline.
  3. Estimate directional trend within each window (increase/decrease/stall).
  4. Flag candidate PSZ where direction reverses between adjacent windows.
  5. Apply falsification gates (windows, thresholds, revision/dataset checks).
  6. Record what survives, what weakens, and what collapses (explicitly).
  7. Generate curiosity prompts that target assumptions—question-only, no answers.

Required outputs

  • PSZ Table: candidate and surviving zones, with evidence of reversal and falsification outcomes.
  • Stress persistence summary: what held, what weakened, what collapsed.
  • Failure modes: limitations of the method (not the domain) stated explicitly.
  • Curiosity prompts (question-only): assumption probes for each surviving PSZ.
  • Map of Ignorance: a visualization that encodes PSZ intensity (no narratives, no causal overlays).

Prohibitions (hard constraints)

  • No causal explanation (no “why this happened”).
  • No forecasting (no “what will happen next”).
  • No policy or recommendations (no actions, interventions, or prescriptions).
  • No moral framing (no blame, praise, or advocacy).
  • No narrative creep: outputs must remain procedural and falsification-driven.

Interpretation boundary

A PSZ is not a claim about reality’s causes. It is a claim about predictive fragility:
that a modeling assumption (often smooth trend continuity) fails to remain stable under reproducible perturbations.
The protocol is designed to mark where certainty breaks—not to replace certainty with a story.

Replicability checklist

  • Data source(s) cited and versioned (including revision year when applicable).
  • Window lengths and thresholds stated explicitly.
  • Directional trend estimator specified (slope/sign method and stall definition).
  • All falsification outcomes recorded (survive / weaken / collapse).
  • Outputs remain question-only (no implied answers).

Page last updated: February 2026

Scroll to Top