High-Impact Domains (Replication Kit)
This page lists twelve high-impact domains where long-horizon predictive assumptions routinely fail.
The Global Curiosity Engine (GCE) does not explain causes, produce forecasts, or recommend actions.
It marks where predictive reliability degrades and returns only disciplined, assumption-focused questions.
How to use this page
- Select one domain from the list below.
- Use the Prompt-Request Prompt to generate a domain-specific Agent Mode prompt (without altering protocol rules).
- Run the resulting prompt in ChatGPT Agent Mode using the stated dataset sources.
- Publish or share outputs as artifacts: PSZ table, persistence summary, collapses list, method failure modes, and question-only prompts.
Note: domain selection does not imply endorsement of a dataset. GCE requires open, citable sources and explicit versioning.
If a domain lacks bulk open data, it should be treated as unsuitable for this protocol until an open series exists.
GCE outputs are expected in the form of artifacts (tables + question prompts), not findings or conclusions
Domain Test Matrix (Choose One)
This matrix is a researcher-facing selection tool. It does not claim what results will be found.
It specifies (a) the baseline predictive assumption to stress-test, (b) public time-series sources,
and (c) why running GCE here is high-leverage for replication and critique.
| # | Domain / Application Area | Core Predictive Assumption Under Test | Key Public Time-Series Sources (2025–2026 updates) | Why GCE Should Be Run Here (Value + Researcher Incentive) |
|---|---|---|---|---|
| 1 | Antimicrobial Resistance (AMR) escalation | Resistance percentages rise gradually/manageably; stewardship bends curves downward predictably. | WHO GLASS dashboard/CSV (≈2016–2025+; annual pathogen–antibiotic pairs by country, where available). | High-stakes forecasting domain with rapidly shifting baselines; useful to test whether “manageable rise” assumptions remain stable under falsification without importing causal narratives. |
| 2 | Biodiversity loss & genetic diversity decline | Species/extinction/genetic loss rates are steady/linear; conservation stabilizes gradually. | IUCN Red List Index (RLI), Living Planet Index (LPI), IPBES/FAO biodiversity-related time-series (availability varies by series). | Measurement and reporting are heterogeneous; GCE can help distinguish method-robust directional breaks from artifacts across revisions and window choices. |
| 3 | Sovereign/public debt sustainability & ratios | Debt-to-GDP converges/stabilizes under growth; shocks are temporary deviations. | IMF World Economic Outlook (WEO) database (annual), World Bank International Debt Statistics (IDS) (annual country-level debt stocks/ratios). | Debt baselines drive planning and risk models; GCE is useful for auditing continuity assumptions under cross-revision checks and perturbations. |
| 4 | Economic inequality (Gini, wealth concentration) | Trends converge/reduce with growth; recovery is gradual after shocks. | World Inequality Database (WID), World Bank inequality indicators (including Gini where available; annual/periodic). | Direction changes are often masked by aggregation and measurement shifts; GCE provides a falsification-first way to test whether “gradual convergence” assumptions hold. |
| 5 | Life expectancy & mortality trend stalls/reversals | Upward convergence is stable; setbacks are transient deviations from a long-run rise. | UN WPP (life expectancy/mortality series), WHO Global Health Observatory (GHO) mortality indicators (annual by country/cause, where available). | Post-shock stalls and reversals raise high-leverage questions about continuity assumptions; GCE can separate method-robust breaks from window/threshold artifacts. |
| 6 | Crop yield & food production under climate stress | Yields continue upward (tech-driven); climate impacts are gradual/manageable. | FAO FAOSTAT (annual crop-specific yields/production by country/region; coverage varies by crop and reporting). | Food systems contain strong non-linearities and regional divergence; GCE can flag persistent directional instability in yield baselines without attributing causes. |
| 7 | Ocean health (acidification / fisheries decline) | Changes are gradual; ecosystems adapt/recover predictably. | GOA-ON / NOAA ocean acidification observations (where open series exist), FAO fisheries capture/effort series (annual; proxy availability varies by region/metric). | Ocean indicators are often sparse and uneven; GCE’s falsification gates help prevent over-claiming while still marking where continuity assumptions fail. |
| 8 | Renewable energy transition & adoption curves | Costs decline smoothly; adoption follows a stable S-curve toward dominance. | IRENA annual cost/capacity/installation series; IEA open indicators where available; national energy statistics (annual). | Regional plateaus and reversals are often treated as “temporary”; GCE can test whether those deviations persist under perturbation and cross-source checks. |
| 9 | Global forced migration & displacement flows | Flows are gradual/directionally stable; components remain predictable. | UNHCR annual displacement datasets; UN DESA migration stocks/flows (annual/periodic; definitional changes documented by source). | Strong definitional and revision sensitivity makes this a good falsification testbed; GCE can map where continuity assumptions break without explaining why. |
| 10 | Inflation persistence & post-shock trajectories | Disinflation is smooth after shocks; inflation returns to target predictably. | IMF inflation series; national statistics agencies / central bank time-series (CPI/inflation; annual/quarterly depending on source). | High policy sensitivity and frequent model error claims make this a valuable audit domain; GCE can locate instability in baseline continuity under explicit thresholds. |
| 11 | Extreme weather event frequency & intensity | Increases are gradual/predictable; impacts reduce smoothly with adaptation. | EM-DAT (where accessible under its terms) and/or national open disaster loss/event datasets; NOAA/other national agencies for event indices where open series exist. | Event series are prone to reporting and definitional artifacts; GCE’s cross-source and perturbation checks are directly useful for separating robust breaks from noise. |
| 12 | AI infrastructure energy demand growth | Demand rises predictably with efficiency gains; grids adapt gradually. | IEA / EIA data center electricity consumption indicators where open; national grid operator statistics (annual/periodic; availability varies). | Planning assumptions are highly sensitive to growth-rate stability; GCE can test whether acceleration patterns persist under perturbation without making forecasts. |
Note: Source availability varies. If a domain lacks a clean, open, citable time-series with versionable downloads,
it is not suitable for strict GCE execution until such a series exists.
GCE outputs are expected in the form of artifacts (tables + question prompts), not findings or conclusions
Step 1 — Use this prompt to generate the domain-specific Agent Mode prompt
Copy/paste the following into a normal ChatGPT chat (not Agent Mode). Replace bracketed fields only.
Do not add goals, conclusions, or interpretations.
ROLE: You are a Prompt Composer for the Global Curiosity Engine (GCE).
TASK: Generate one executable ChatGPT Agent Mode prompt that applies the GCE protocol to the selected domain.
The prompt must be standalone and must include all constraints, falsification gates, and required outputs.
SELECTED DOMAIN: [PASTE ONE DOMAIN FROM THE LIST]
PRIMARY DATA SOURCE (OPEN): [PASTE THE DATASET NAME + LINK OR DOWNLOAD PAGE]
VARIABLE(S): [E.G., TFR, CPI, TEMPERATURE, RESISTANCE %, ETC.]
GEOGRAPHIC UNIT: [COUNTRY / STATE / REGION / OTHER]
TEMPORAL SCOPE: [E.G., 1950–PRESENT]
NON-NEGOTIABLE CONSTRAINTS:
- Act as an independent execution agent for GCE (not a scientist, not an explainer).
- Do not interpret causes, propose solutions, make forecasts, or recommend actions.
- Primary stress signal: TREND INSTABILITY ONLY (directional reversals across rolling baselines).
- Must include falsification gates:
1) Window perturbation (e.g., 20y vs 30y or equivalent)
2) Threshold sensitivity (stall vs increase/decrease thresholds)
3) Cross-revision or cross-dataset consistency when available
- Must explicitly list signals that collapse under falsification.
REQUIRED OUTPUT SECTIONS (EXACT):
1) Stress Persistence Summary
2) Zone Stability Assessment (Aggregate, not exhaustive)
3) Signals That Collapsed (Explicitly listed)
4) Failure Modes of the Method (Not the Domain)
5) Confidence in the GCE Instrument (Not in domain forecasts)
6) Curiosity Prompts (Question-Only)
STOP CONDITION:
Stop immediately after delivering the required output sections. No additional commentary.
DELIVERABLE:
Return ONLY the final Agent Mode prompt text (no analysis, no alternatives).Step 2 — Run the generated prompt in ChatGPT Agent Mode
Once the domain-specific Agent Mode prompt is generated (Step 1), run it in a new ChatGPT discussion using Agent Mode.
Provide the dataset link(s) and ensure the agent can access the data (CSV downloads preferred).
Step 3 — What outputs to expect (artifact expectations only)
The researcher should expect the following deliverables, regardless of domain. These are not conclusions about reality;
they are artifacts that describe predictive fragility under falsification.
- Stress Persistence Summary: A high-level description of whether directional reversals persist after perturbations.
- Zone Stability Assessment: A compact table listing candidate PSZ and whether they survive/weaken/collapse.
- Signals That Collapsed: An explicit list of candidates that fail under falsification (required for credibility).
- Failure Modes (Method): A candid section describing window/threshold/dataset limitations and artifacts.
- Instrument Confidence (Not Forecast Confidence): A bounded statement about whether the protocol behaved as intended.
- Curiosity Prompts (Question-Only): Assumption-probing questions for each surviving PSZ (no answers).
Important boundary
Expected outputs describe format and procedural behavior only. This page does not claim what any researcher will find
in any specific country, region, or dataset. GCE is explicitly designed to avoid narrative substitution.
Page last updated: February 2026