Apocalypse Clock is a transparent, scenario-based systemic-risk model. Its purpose is to make interacting global risks comparable within a single analytical framework: it examines 23 global threats across civilizational, biospheric, and technological domains, not as isolated dangers, but as coupled components of an interconnected global system.
Its parameter data are drawn primarily from official governmental reports, intergovernmental and international-organization assessments, scientific literature, and evidence-based institutional datasets. Each parameter is documented through source references, uncertainty ranges, evidence strength, growth calibration, and explanatory notes.
The model estimates a probabilistic critical horizon through a multi-stage analytical pipeline: evidence-graded parameter scoring, Weighted MCDA, dependency amplification, domain-weight normalization, process-specific horizon functions, Monte Carlo uncertainty sampling, bootstrap uncertainty estimation, structural ensemble aggregation, and Dynamic cascade propagation.
The ensemble compares compensatory aggregation, non-compensatory max-rule aggregation, graph-weighted aggregation, and Dynamic cascade logic. The paired highlighted years represent the P50 median and P90 later edge of the model’s Dynamic cascade horizon. Each displayed year is a model-derived threshold horizon under explicit assumptions: a stress-oriented critical threshold, not a deterministic forecast, a prophecy, or an empirically measured probability of collapse.
The optional Model Diagnostics adds deeper diagnostic tools, including Weibull survival analysis, network eigenvector centrality, Poisson-binomial convergence-tail analysis, Shannon entropy, Fast OAT sensitivity analysis, Sobol/Jansen sensitivity indices, SMAA weight robustness, non-compensatory veto diagnostics, tail-dependence stress testing, and a scientific audit summary. These diagnostics explain and stress-test the baseline result; they do not overwrite the paired headline horizons.
The expanded view below breaks that horizon into the model steps behind it: threat scoring, dependency links, process-specific horizons, uncertainty sampling, aggregation rules, and cascade diagnostics.
Results remain sensitive to thresholds, dependency strength, domain weights, growth assumptions, and aggregation rules. Optional diagnostics make that sensitivity visible; they are interpretive tests, not a second headline forecast.
Share of simulated runs in which the functional-disruption threshold has been reached by each year. It describes the model under its stated assumptions about growth, thresholds, weights and connections. It is not a measured probability and not a forecast.
This panel separates the model into two computational layers. The core run produces the displayed horizon from evidence-informed scoring, dependency amplification, process-specific horizon functions, Monte Carlo uncertainty sampling, structural ensemble aggregation, and dynamic cascade testing. The Model Diagnostics adds an optional diagnostic layer after the baseline result exists. It visualizes sensitivity, weight robustness, non-compensatory stress rules, correlated tail shocks, and the distributional structure of the current Monte Carlo result. These diagnostics are for auditability and interpretation under deep uncertainty; they do not overwrite the headline clock date.
Each threat receives six dimension scores. The model combines them into the same weighted baseline score used by the ranking code, then applies scenario, domain weighting and dependency effects.
Threats are not treated as independent. A threat can amplify another through declared dependency links, producing the same adjusted priority value used for ranking and horizon calculation.
The model does not convert all risks into time the same way. Continuous, event-like and regime-shift threats use the process-specific horizon functions implemented in the engine.
Uncertain inputs are sampled repeatedly. The output is summarized through P10, P50 and P90 rather than a single deterministic year.
Different aggregation rules answer different questions: weighted mass crossing, earliest single threat, graph-linked joint stress and dynamic cascade propagation.
A single initial functional failure can spread through directed dependencies. The trigger uses criticality weights globally or within an essential-service basket, and each run samples the threshold and the other structural choices; neither extinction of all organisms nor initial failure in every domain is required.
OAT shows local one-at-a-time leverage. Sobol/Jansen estimates direct and total variance contribution. SMAA perturbs weights to test whether top-ranked drivers remain stable.
Veto and tail-dependence diagnostics test non-compensatory and correlated-stress alternatives. Histogram, boxplot, heatmap and CDF views make the baseline uncertainty structure visible.
| # | Threat | Sc | Ur | Acc | Int | Irr | Gov | Base score | Adj. score | Model Ti | Model P10–P90 | Effective risk-growth proxy | Model threshold | Scientific evidence | Input calibration | Mechanism |
|---|
