The Test of Truth: Computational Falsifiability (Axiom A8)

Published: February 28, 2026
Read time: 10 minutes
Category: Axioms Series

What separates science from storytelling? Both explain the world. Both offer insights. Both can be compelling. But there’s a crucial difference: science can be wrong. A scientific theory makes predictions that could fail. When predictions fail repeatedly, the theory is rejected. This vulnerability to refutation—falsifiability—is what distinguishes science from speculation. Axiom A8 ensures that Computational Macrohistory meets this standard.

The Requirement

Every CMH model must generate predictive distributions—P_M(X(t+Δt) | X(t))—that can be quantitatively compared with observed data. Not vague claims that fit any outcome, but specific, temporal, observable predictions that could be proven wrong.

In This Post

The full article explores:

  • The demarcation problem: Why falsifiability matters for scientific status
  • What makes predictions testable: Specificity, temporality, observability, independence
  • Metrics of falsification: Brier Score, AUC-ROC, calibration plots
  • The backtesting protocol: Train/test splits, cross-validation, no data leakage
  • Publishing failures: Why transparency about errors builds credibility
  • Computational reproducibility: Open code, accessible data, replicable results
  • Completing the foundation: How A1-A8 together define CMH as a science

With Axiom A8, the theoretical foundation is complete. The empirical work begins.

📖 Read the full article:
👉 The Test of Truth: Computational Falsifiability (Axiom A8)

💬 Join the discussion:
Comments welcome on Substack

📧 Subscribe for weekly updates:
Computational Macrohistory Bulletin

Previous: The Horizon of Foresight: Predictive Decay (Axiom A7)
Next: Applying the framework—case studies in computational macrohistory

Similar Posts