Tier9
Research-Backed

Our Methodology

Tier9 is built on decades of research in superforecasting, prediction markets, and decision science. Here's how we turn uncertainty into actionable intelligence.

Interactive

AI-Human Ensemble in Action

Adjust the weights to see how different combinations affect the ensemble forecast.

Ensemble Forecast
70.1%
Weighted combination of all sources
Claude 4.5
72%
Weight: 30%
GPT-5.1
68%
Weight: 25%
Gemini 3 Pro
74%
Weight: 25%
Human Crowd
65%
Weight: 20%
Total weight: 100%
Pro users can customize AI weights for their organization
Upgrade
Advanced

Extremization: Reducing Groupthink

When forecasters share information, the crowd consensus often becomes too moderate. Extremization corrects for this.

Raw Crowd Consensus
65%
Extremized Forecast
73%
Extremization Factor1.5x
1.0x (no change)3.0x (strong extremization)

Why Extremize?

Research from the Good Judgment Project shows that when forecasters discuss and share information, they often converge toward moderate probabilities. Extremization pushes forecasts away from 50% to account for this shared information effect, improving accuracy.

Core Principles

Our approach is grounded in peer-reviewed research from the Intelligence Advanced Research Projects Activity (IARPA), Good Judgment Project, and leading academic institutions.

Proper Scoring Rules

Unlike some platforms that use improper scoring, Tier9 uses the Brier score—a strictly proper scoring rule that incentivizes honest probability reporting.

Brier Score = (forecast - outcome)²

0 = perfect prediction, 1 = worst possible prediction

Calibration Training

Research shows that calibration training can reduce overconfidence and improve forecast accuracy. Our platform provides real-time calibration feedback.

  • Interactive calibration charts
  • Personalized AI coaching
  • Benchmark comparisons
Crowd Wisdom Aggregation

We combine multiple aggregation methods to extract signal from diverse forecaster populations:

  • Median: Robust to outliers
  • Weighted average: Track record based
  • Extremized mean: Reduces groupthink
AI-Human Ensemble

We combine human forecasts with AI model predictions to create ensemble forecasts that outperform either alone:

Claude 4.5 SonnetActive
GPT-5.1Active
Gemini 3 ProActive
DeepSeek R2Active

Experience Our Methodology

See how our forecasting system works with real questions and live probabilities.