Prediction Market
ProMultiple AI models provide probability estimates for future events with confidence intervals, key factors, and a statistical consensus analysis.
Overview
Prediction Market mode treats each AI model as an independent forecaster. Each model provides a probability estimate (0-100%) for your prediction question, along with factors that could push the probability higher or lower. The system then calculates statistical consensus including mean, median, and standard deviation to show agreement levels.
Key Features
Probability Estimates
Each model provides a 0-100% probability with confidence intervals (e.g., 65% ± 15%).
Web Search Tool
Models can search the web for current information, news, and data to inform predictions.
Factor Analysis
Each model identifies factors that could push probability higher or lower, plus biggest uncertainties.
Statistical Consensus
Mean, median, standard deviation, and agreement level calculated from all model estimates.
How It Works
Question Formulation
Submit a prediction question with a resolution date. The question should be specific and have a clear yes/no outcome.
Parallel Forecasting
ParallelAll models generate predictions simultaneously. Each may use web search to gather current information before making their estimate.
Statistical Consensus
Once all estimates are collected, the system calculates mean, median, and standard deviation. Agreement and disagreement points are identified by comparing factors across models.
What Each Model Provides
The model's probability estimate that the predicted event will occur by the resolution date.
How certain the model is about its estimate. A smaller interval means higher confidence.
Key factors that could push the probability higher or lower than the estimate.
The single most important unknown factor that could significantly change the outcome.
Example Output
Prediction Question
"Will GPT-5 be released before July 2026?"
Resolution: July 1, 2026
"Based on OpenAI's historical release cadence and recent announcements, likely but not certain..."
"Significant uncertainty given competitive landscape and safety considerations..."
"Industry pressures suggest a release but technical challenges may cause delays..."
Agreement: All models cite competitive pressure and historical cadence as positive factors.
Disagreement: Models differ on impact of safety concerns and regulatory environment.
When to Use Prediction Market
API Usage
Prediction Market mode uses the Konnect OpenAI-compatible API with pattern extensions.
curl https://api.konnect.ai/v1/chat/completions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "konnect-prediction",
"messages": [
{"role": "user", "content": "Will SpaceX land humans on Mars before 2030?"}
],
"stream": true,
"konnect.pattern": "prediction",
"konnect.models": ["gpt-5.2", "claude-sonnet-4-5-20250929", "gemini-2.5-flash"],
"konnect.prediction_resolution_date": "2030-01-01"
}'The models make independent probability estimates, and the system calculates statistical consensus. Each model has access to web search for gathering current information.
Streaming Events
Prediction Market mode streams events as each model completes its estimate:
estimate_completeSent when a model completes its probability estimate. Includes model name, probability, confidence interval, factors, and biggest uncertainty.
consensus_completeSent when all estimates are collected and statistical consensus is calculated. Includes mean, median, standard deviation, range, and agreement/disagreement points.
{
"konnect.metadata": {
"prediction_event": "estimate_complete",
"estimate": {
"model": "gpt-5.2",
"probability": 72,
"confidence_interval": 12,
"factors_higher": [
"Competitive pressure from other labs",
"Strong financial position"
],
"factors_lower": [
"Safety evaluation delays",
"Regulatory scrutiny"
],
"biggest_uncertainty": "Compute availability and cost"
}
}
}Agreement Levels
Agreement level is determined by the standard deviation of probability estimates:
Models are closely aligned in their probability estimates.
Models have moderate disagreement on probability.
Models significantly disagree. Consider this high uncertainty.