Our Edge Methodology
How Foresight finds tradeable opportunities in prediction markets.
The Core Thesis
Prediction markets are priced by crowds, but different crowds on different platforms see different information at different speeds. When the same event is traded on Polymarket, Kalshi, Metaculus, and Manifold, price disagreements between platforms represent either:
- Informational edge — one platform's participants have faster or better information
- Structural mispricing — liquidity differences, fee structures, or demographic biases create persistent gaps
- Repricing lag — breaking news reaches one platform before others
Foresight systematically identifies these disagreements and enriches them with news sentiment to determine which side is likely correct.
Signal Types
1. Cross-Platform Divergence
The primary signal. When the same event shows a probability spread of more than 5 percentage points across platforms, we flag it. The wider the spread, the higher the signal score.
Platform accuracy weights are derived from historical resolution data — platforms that have been more calibrated on similar event types get higher weights.
2. Momentum Detection
We take snapshots at regular intervals and track probability velocity. A market moving more than 3 percentage points in 24 hours with increasing volume signals a repricing event. Momentum is most actionable when only one platform is moving while others are static.
3. Consensus vs. Outlier
When three platforms agree within 2 points and one disagrees by 8+, the outlier is either ahead of the curve or wrong. News enrichment helps determine which. If recent headlines support the outlier's implied direction, we weight it as a leading signal.
4. Cross-Market Correlation
Some events are logically linked — if a ceasefire becomes likely, defence stocks and crypto markets may react. We track correlation coefficients between event clusters to identify second-order opportunities that the market has not yet priced in.
News Enrichment
Every signal is enriched with real-time news via Brave Search. We perform sentiment analysis on headlines related to each event, scoring them from -1 (bearish) to +1 (bullish). When market price and news sentiment diverge, the tradeable opportunity is usually in the direction of the sentiment — markets lag narrative by 6-48 hours on average.
Position Sizing
We use a modified Kelly Criterion for position sizing recommendations:
Where:
edge= our estimated true probability minus the market priceodds= implied odds from the market priceconfidence_discount= a conservative multiplier (typically 0.25-0.5x Kelly) to account for model uncertainty
We never recommend risking more than 5% of bankroll on a single position, regardless of calculated edge.
Historical Accuracy Tracking
Every resolved market is logged and scored against our predictions. We track calibration by source, category, and signal type. This feedback loop continuously improves platform accuracy weights and signal thresholds. Current resolution history is limited but growing with each market cycle.
Data Sources
- Polymarket — Gamma API + CLOB for real-time order book data
- Kalshi — Public trade API with event-level aggregation
- Metaculus — Authenticated API for community forecasts and question metadata
- Manifold — Public API for market probabilities and resolution data
- Brave Search — News enrichment and sentiment analysis
Limitations
Prediction markets are not efficient in the academic sense, but they are getting better. Our edge is in the speed and breadth of cross-platform analysis, not in fundamental forecasting superiority. As markets mature and arbitrage bots improve, divergences may shrink. We adapt by continuously refining our signal thresholds and expanding to new data sources.
Past signals are not guarantees. This is intelligence, not financial advice.