Documentation

Sign up at polyedgeai.com/dashboard using your wallet or email. Your Starter plan activates immediately with 2 AI market analyses and 1 deep research report per day. No credit card required.

Wallet connection is optional and only for users who want to track their own Polymarket positions within the PolyEdge AI dashboard. We never have access to your funds — the connection is read-only and only used to display your existing positions alongside our AI analysis. You can use all research features without connecting a wallet.

The dashboard is your research home base. It shows active markets with AI probability forecasts, model agreement scores, and whale activity indicators. Use the category tabs to filter by Politics, Crypto, Sports, Economics, Tech, Culture, Science, or Other. Each market card shows our AI probability, the current market price, and the analytical gap between them.

Click any market to open its full analysis report. You'll see: (1) The AI probability estimate and model agreement score, (2) Key factors driving the forecast, (3) Data quality assessment, (4) Sentiment analysis from social sources, (5) Whale activity context, and (6) The detailed reasoning from our 6-model ensemble. Each report takes approximately 2 seconds to generate.

Our prediction engine runs six frontier AI models in parallel on every market analysis: • Claude Sonnet — Anthropic's strongest reasoning model, our anchor for complex markets • Claude Haiku — Faster Anthropic variant providing an independent estimate • GPT-4o — OpenAI's flagship, strong on technology and science • GPT-4o-mini — Lighter variant whose divergence from GPT-4o is itself a signal • Perplexity Sonar — Real-time web search grounded in current data • Grok — xAI's model with live social sentiment from X Each model independently estimates a probability. We then compute a weighted average (weights based on historical category accuracy), apply disagreement scoring, favourite-longshot bias correction, and per-category calibration multipliers to produce a single calibrated probability.

Continuous Ranked Probability Score (CRPS) is the gold standard for evaluating probabilistic forecasters. Unlike simple win/loss tracking, CRPS measures how close your probability estimate was to the actual outcome. A forecast of 90% on an event that happens scores much better than a forecast of 55% on the same event — even though both are technically "correct." This incentivizes precision, not just being on the right side. Lower CRPS is better. Our current average CRPS across all categories is 0.127. We recalibrate per-category multipliers nightly based on CRPS scores from resolved forecasts.

The "edge" or analytical gap represents the difference between our AI's calibrated probability and the current market price. For example, if our AI estimates 75% and the market is priced at 62%, the analytical gap is +13 percentage points. A larger gap suggests the market may be mispricing the event relative to our AI's assessment. However, the AI can be wrong — and frequently is on certain types of markets. The gap is most informative when model agreement is high (5/6 or 6/6) and the category has strong historical accuracy. Always evaluate the reasoning behind the forecast, not just the number.

Model agreement indicates how many of our 6 models directionally agree on the forecast. "6/6 models agree" means all models are on the same side and tightly clustered. "3/6" means the models are split — a sign of genuine uncertainty. High agreement (5/6 or 6/6) historically correlates with higher forecast accuracy. Low agreement (3/6 or 4/6) often indicates markets where the outcome is genuinely uncertain and our confidence is lower. We display model agreement alongside every forecast so you can factor it into your own research.

Our whale detection system monitors Polymarket's CLOB (Central Limit Order Book) for unusual activity patterns including large-block orders, rapid accumulation, and order book imbalances. When a large player places significant capital on one side of a market, we flag it as whale activity and show whether it agrees or disagrees with our AI forecast. This is contextual information — whales can be right or wrong — but it's a useful additional signal for research.