Our Methodology
A deep dive into how we create, track, and update probabilistic predictions using AI-powered signal analysis.
Base Rate Selection
Every prediction starts with a base rate - our initial probability estimate before any news signals are applied. Base rates are determined by:
- Historical reference classes: What happened in similar situations?
- Expert consensus: What do domain experts believe?
- Market signals: What do prediction markets suggest?
- Structural factors: What are the inherent constraints?
We document our base rate justification for every prediction, making our initial reasoning transparent.
Signal Detection
Our AI system monitors news sources daily to extract signals - pieces of information that affect prediction outcomes.
Increases probability of YES outcome
Decreases probability of YES outcome
Context without clear directional impact
No relevant news (can be informative)
Each signal is assigned a strength from 0-100, representing its potential impact on the prediction.
Probability Update
Signals are applied to update probabilities with built-in safeguards:
Daily Caps
- • Maximum +15% increase per day
- • Maximum -20% decrease per day
- • Probability range: 1% to 99% (never certain)
These caps prevent overreaction to single news events and ensure stable, well-calibrated forecasts.
Resolution & Learning
When a prediction resolves, we conduct a post-mortem analysis:
- Was the base rate appropriate?
- Which signals were real vs. noise?
- What did we miss?
- What can we learn for future predictions?
See It In Action
Explore our predictions to see this methodology applied to real-world events.
View Predictions