What about domain Performance?

| Domain | Accuracy | Calibration Error | Total Predictions | |---|---|---|---| | Geopolitics | 49.5% | 0.3907 | 2,037 | | Markets | 56.4% | 0.4033 | 2,532 | | Other | 77.7% | 0.3546 | 18,220 | | Technology | 47.0% | 0.3394 | 888 | | Energy | 51.9% | 0.2203 | 1,179 | | Defense | 54.5% | 0.5079 | 66 |

AI Prediction Accuracy Report — June 2026

Executive Summary

Quantitative modeling achieved 70.9% accuracy across 24,922 resolved predictions in June 2026, with a probability calibration error of 0.3453 (-9.5 percentage points better than random). Forecasting markets significantly overestimated cryptocurrency price stability, with three of the top five misses involving Ethereum and Bitcoin price thresholds.

Domain Performance

Domain	Accuracy	Calibration Error	Total Predictions
Geopolitics	49.5%	0.3907	2,037
Markets	56.4%	0.4033	2,532
Other	77.7%	0.3546	18,220
Technology	47.0%	0.3394	888
Energy	51.9%	0.2203	1,179
Defense	54.5%	0.5079	66

Geopolitics: Underperformed random chance (49.5% accuracy) with the second-worst calibration error (0.3907). Models systematically overestimated the likelihood of diplomatic breakthroughs.

Markets: Delivered marginal predictive value (56.4% accuracy) but suffered from severe overconfidence in cryptocurrency price stability, evidenced by the highest domain calibration error (0.4033).

Technology: The lowest accuracy domain (47.0%) with moderate calibration issues. Forecasting struggled with emergent technology adoption timelines and regulatory outcomes.

Calibration Analysis

Quantitative modeling exhibited systematic miscalibration across probability bins. The most severe overconfidence occurred in the 90-100% bin, where predictions averaging 94% confidence materialized only 9.1% of the time. Underconfidence appeared in the 20-30% bin, where events predicted at 24% likelihood actually occurred 52.3% of the time. Energy forecasts showed the best calibration (0.2203 error), while defense predictions had the worst (0.5079).

Notable Calls

Top Hit: Temperature predictions for Hong Kong and Wellington demonstrated perfect accuracy (100% confidence, 100% occurrence) across multiple identical forecasts, reflecting high reliability in short-term meteorological modeling.

Market Timing: Correctly predicted Bitcoin's dip to $73,000 on June 1 with 100% confidence, capturing precise intraday volatility patterns through liquidity flow analysis.

Cryptocurrency Misses: Three separate 99%-confidence Ethereum price predictions failed when the asset collapsed below $1,400. Models overweighted technical support levels and underweighted exchange liquidity crises.

Geopolitical Overreach: The 99%-confidence prediction of a US-Iran peace deal by July 31 ignored hardening positions in Tehran's nuclear negotiation team, demonstrating inadequate real-time factional analysis.

Methodology

We evaluate forecasting accuracy by comparing predicted probabilities against binary outcomes. Each prediction receives a correctness score (1 for correct, 0 for incorrect) and a calibration score measuring the deviation between predicted and actual event frequencies. The probability calibration error aggregates these deviations across all predictions, with 0 representing perfect calibration and 0.25 representing random chance. We exclude predictions with less than 24-hour resolution windows from scoring.

Looking Ahead

June's results indicate structural weaknesses in high-confidence market predictions and geopolitical assessments. The 9.1% realization rate for 90-100% confidence calls demands immediate confidence threshold recalibration. Energy forecasting's strong performance (0.2203 error) suggests replicable methods for other commodity domains. Defense and technology predictions require enhanced scenario testing to address chronic underperformance.