How the EV Scanner Works

The EV Scanner is the core intelligence tool in Predite. It continuously monitors prediction markets across Polymarket and Kalshi to identify opportunities where the market price diverges meaningfully from estimated true probability. Both venues are scanned on every plan — the POLY/KLSH chips filter what is already there rather than unlocking anything.

This document covers how the scanner works under the hood, how to interpret its output, and how to integrate it into your trading workflow.

What the Scanner Actually Does

Every 4 hours, our infrastructure:

1. Fetches the current list of active markets from Polymarket and Kalshi, in parallel
2. Pulls orderbook data, volume, liquidity, and recent price action
3. Sends market questions and context through our 3-model AI consensus engine
4. Computes expected value for each potential position
5. Ranks markets by absolute edge size
6. Filters out markets that fail liquidity, freshness, or quality checks

You see the results in the scanner table. Click any column header to sort by it — first click descending, second ascending, a third returns to the default order. That default is absolute edge, descending, and it matters: a strong NO at −18pp outranks a weak YES at +6pp, so the top of the table is the biggest disagreement with the market, not the best YES.

Reading the Scanner Table

The default columns are:

Market: question being asked (e.g., "Will Trump win the 2026 election?")

Market %: current market price expressed as implied probability. A YES contract at $0.40 = 40% implied probability.

AI Est.: our AI's estimated probability based on news, polling, base rates, and reasoning. Range 0-100%.

Edge: difference in percentage points (pp) between AI estimate and market price. Positive edge = AI thinks YES is mispriced low (buy YES). Negative edge = AI thinks YES is mispriced high (buy NO).

Signal: simplified action recommendation. BUY YES, BUY NO, or HOLD (edge too small or confidence too low).

Conf.: AI confidence level 0-100%. Higher = more reliable estimate.

Volume: the market's cumulative trading volume in USD, over its whole life — not the last 24 hours. Useful as a read on total attention, useless as a read on what is happening today.

Resolution date is not shown. The scanner does not have a Resolution column, and the market's end date is not even sent to the table. Time to resolution matters — shorter horizons carry less variance, longer ones tie up capital — but you have to open the market to find out. The Resolution Tracker is the page built for that view.

What Edge Sizes Mean

Based on backtesting and real-world experience:

0-3pp edge: skip. Noise floor — below this, slippage and fees consume any theoretical edge.

3-5pp edge: marginal. Worth investigating but trade small. Most edges below 5pp don't survive execution costs.

5-10pp edge: real signal. Most of your profitable trades will be in this range. Position size 2-5% of bankroll.

10-15pp edge: strong signal but verify. Verify resolution criteria, check news, confirm liquidity.

>15pp edge: rare and suspicious. Usually means you're missing something or the market is illiquid. Re-check before trading.

What Confidence Means

The AI provides a confidence level alongside its estimate:

70-100% confidence: high. AI has multiple converging signals and clean reasoning.

50-70% confidence: medium. Some uncertainty in the analysis. Smaller positions advisable.

<50% confidence: you will not see this. Every model in the engine floors its own confidence at 50 (Claude is clamped to 50–95, the heuristic to 50–90, and the microstructure model starts at 55 and only climbs), so the weighted consensus cannot land below 50 either. Read a 50–60 score as "the engine has little to say here", not as a rare low-confidence warning.

For most trades, target signals with confidence >70%. Below that, you're amplifying AI's uncertainty.

Filtering and Sorting

The scanner supports several filters:

That is the entire filter bar. There is no minimum-volume filter, no maximum-days-to-resolution filter and no adjustable edge threshold — the 5pp chip is the only edge control, and it does not slide.

Common Scanner Patterns

After using the scanner for a while, you'll notice patterns:

Major elections: Polymarket has very efficient pricing on US presidential markets. AI rarely shows significant edge. Trade only with specific informational advantage.

Niche politics (state-level, congressional): more inefficient. Often 5-10pp edges available. Requires research to verify.

Crypto markets: high variance. Big edges sometimes, but resolution risk is also higher.

Sports: depends on the sport. Mainstream US sports (NFL, NBA) very efficient. International soccer, niche events more inefficient.

Economic indicators: extremely efficient on Kalshi. Skip unless you have specific data sources.

How to Use Scanner Output

A reasonable workflow:

1. Open scanner first thing in the trading day
2. Filter for >5pp edge, >70% confidence, >$10k volume
3. Read top 10-15 results
4. Pick 2-3 that interest you (you can articulate WHY there's edge)
5. Click into each for full AI reasoning + market context
6. Verify resolution criteria, recent news, liquidity
7. If still convinced, place trade with appropriate sizing
8. Track each trade in your journal

This takes about 20-30 minutes daily. After 100+ trades using this workflow, you'll know whether your edge identification is accurate enough to compound.

Limitations to Know

The scanner is a starting filter, not a magic oracle:

Always verify scanner output with your own analysis. The scanner saves you time by filtering thousands of markets to a manageable list — but YOU make the final trade decision.

Customization for Power Users

Saved filter presets do not exist on any plan. The scanner's filters are client-side state and reset when you navigate away.

Related Docs