GJP foundations
Mellers et al. (2014), Psychological Strategies for Winning a Geopolitical Forecasting Tournament, Psychological Science, 25(5), 1106–1115. Publisher. It reports effects of probability training, collaboration and selecting strong performers in a tournament.
Mellers et al. (2015), Identifying and Cultivating Superforecasters as a Method of Improving Probabilistic Predictions, Perspectives on Psychological Science, 10(3), 267–281. Publisher. It examines persistent high performance and several associated explanations.
Chang, Chen, Mellers and Tetlock (2016), Developing Expert Political Judgment: The Impact of Training and Practice on Judgmental Accuracy in Geopolitical Forecasting Tournaments. Journal. It studies training and practice within this research tradition.
Evidence is not unanimous
Hauenstein, Thomas, Illingworth and Dougherty (2025; first online 2024), Rethinking the Role of Teams and Training in Geopolitical Forecasting: The Effect of Uncontrolled Method Variance on Statistical Conclusions. Publisher.
This reanalysis challenges interpretations of latent ability after accounting for method variables. It does not erase every observed performance difference or constitute a new prospective replication. Do not present causal explanations as beyond dispute.
Attribution discipline
- Credit Bayes/Price for the historical inverse-probability foundation, not Tetlock.
- Credit Brier for the score and Murphy for its classic decomposition.
- Credit Kahneman/Tversky for early corrective outside-view work and Flyvbjerg for project applications.
- Credit Dalkey/Helmer for the named Delphi method.
- Credit Lord/Lepper/Preston for the named consider-the-opposite intervention.
- Credit Satopää and coauthors for the specified logit aggregator.
- Credit McCaslin and coauthors for FRI’s named conditional-tree elicitation method.
- Treat versioned schemas, code helpers and AI safeguards as this pack’s adaptations.
Each standalone technique contains fuller provenance. Historical ideas are often cumulative; where no sole inventor is established, say so. Do not fabricate a first-inventor claim.
What the pack does not establish
It does not reproduce proprietary Good Judgment services, confer Superforecaster status, promise superior forecasts, or establish that human training effects transfer to AI. It does not contain full copyrighted papers. It provides original instructions, mathematical implementations and linked research.
Research grounding and Anthropic format compatibility are different claims. Neither implies actual Claude runtime acceptance or measured real-world forecasting accuracy.