Edge Detection Engine

01 — THE MATH

Market price
vs true probability
= your edge.

Edge is simple. It's the difference between what the market is priced at and what the AI estimates the true probability should be. Market says 40%. AI says 68%. Edge: +28%. Buy YES. No edge? No action. HOLD. Every market on Polymarket gets an edge score, a confidence level (HIGH / MEDIUM / LOW), and a recommendation.

—

Edge Calculation

Edge % = (AI estimate - Market price) / Market price × 100. Simple math. The larger the edge, the more actionable the signal.

—

Confidence Scoring

HIGH: Multiple data sources agree, past accuracy is strong, no calibration red flags. MEDIUM: Some disagreement, moderate past accuracy. LOW: Weak signals or poor historical performance on this topic.

—

No-Edge HOLD

If edge < 5% or confidence is LOW, the recommendation is HOLD. There's no actionable edge. No trade = no risk. This is the safety valve.

polytragent — edge scoring

polytragent — confidence assessment

02 — THE SCORING

How does
Polytragent
assign
confidence?

Confidence is multifactorial. It's a weighted average of multiple signals: sub-market agreement, Manifold cross-reference validation, prediction history on this topic, overall calibration, recency (how fresh are the underlying data points). A 20% edge with LOW confidence is ignored. A 10% edge with HIGH confidence is a BUY signal.

Sub-Market Agreement

If multiple Polymarket sub-markets on the same event all suggest the same direction, confidence goes up +20%.

Manifold Cross-Reference

If Manifold forecasters agree with the AI's edge direction, add +25% to confidence. If they disagree, subtract -15%.

Topic-Specific Accuracy

If the AI has >80% accuracy on this topic category, add +15%. If <65%, subtract -20%.

Calibration Score

Is the AI over/underconfident on this kind of market? If perfectly calibrated, no adjustment. If biased, reduce confidence by the bias amount.

Data Freshness

Older market data gets a freshness penalty. Markets with volume updates in last 2 hours: no penalty. Stale data (>24h): -10% confidence.

Final Confidence Bucket

Calculate weighted score. Round to HIGH (>75%), MEDIUM (50–75%), or LOW (<50%). Display recommendation.

Edge detection
FAQ

What's the minimum edge to act on?

Edges under 5% are noise. After transaction fees (~2%), you're breaking even. Polytragent flags 5–10% edges as MEDIUM confidence, 10%+ as potential HIGH confidence signals.

How does Polytragent handle disagreement between sub-markets?

If two sub-markets suggest opposite directions, the AI weights them by liquidity. The sub-market with higher trading volume gets more vote. This reduces false signals.

Can I trade on LOW confidence edges?

You can, but Polytragent doesn't recommend it. LOW confidence recommendations are flagged as HOLD. The reasoning: expected value is negative after factoring in risk.

What if the market moves before I execute?

Edge scores refresh on every /research command. If the market price has moved closer to the AI's estimate, edge decreases. This is live data.

How accurate is the AI's true probability estimate?

Over the full year of data, the AI's calibration is 78% accurate. This varies by topic: 85% on economics, 60% on crypto. These biases are explicitly tracked and accounted for.

What if no data sources agree?

Confidence is capped at MEDIUM or LOW. Polytragent will still recommend action if the edge is >15%, but with transparency that underlying confidence is weak.

Find edges
with confidence.

Every recommendation comes with a confidence score. HOLD when there's no edge. BUY when it's real.

Start trading with confidence — $79.99/mo →