Prediction Market

Pro

Multiple 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.

Estimates
Web Search
Statistics
Consensus

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

1

Question Formulation

Submit a prediction question with a resolution date. The question should be specific and have a clear yes/no outcome.

2

Parallel Forecasting

Parallel

All models generate predictions simultaneously. Each may use web search to gather current information before making their estimate.

3

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

Probability (0-100%)

The model's probability estimate that the predicted event will occur by the resolution date.

Confidence Interval (± %)

How certain the model is about its estimate. A smaller interval means higher confidence.

Factors Higher/Lower

Key factors that could push the probability higher or lower than the estimate.

Biggest Uncertainty

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

GPT-5.272% ± 12%

"Based on OpenAI's historical release cadence and recent announcements, likely but not certain..."

Claude Sonnet 4.565% ± 18%

"Significant uncertainty given competitive landscape and safety considerations..."

Gemini 2.5 Flash58% ± 20%

"Industry pressures suggest a release but technical challenges may cause delays..."

Consensus
65%
Mean
65%
Median
7.0
Std Dev
High
Agreement

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

Technology and product launch predictions
Market and economic forecasting
Sports and event outcome predictions
Policy and regulatory change forecasting
Scientific and research milestone predictions
Any yes/no question with a future resolution date

API Usage

Prediction Market mode uses the Konnect OpenAI-compatible API with pattern extensions.

cURL
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_complete

Sent when a model completes its probability estimate. Includes model name, probability, confidence interval, factors, and biggest uncertainty.

consensus_complete

Sent when all estimates are collected and statistical consensus is calculated. Includes mean, median, standard deviation, range, and agreement/disagreement points.

Example estimate_complete event
{
  "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:

High AgreementStandard deviation < 10

Models are closely aligned in their probability estimates.

Medium AgreementStandard deviation 10-20

Models have moderate disagreement on probability.

Low AgreementStandard deviation > 20

Models significantly disagree. Consider this high uncertainty.

Explore other modes