1The basics: YES + NO should equal $1.00

The concept is dead simple. On Polymarket, every market has a YES token and a NO token. They should always add up to $1.00. If YES is $0.60, NO should be $0.40. That's the law.

But they don't always add up to $1.00.

Example: Simple YES/NO Arbitrage

Buy YES

$0.60

Buy NO

$0.35

Total Cost

$0.95

Guaranteed Profit

+$0.05 (5.3%)

That's arbitrage. And while 5% might not sound exciting, when you can find these dozens of times a day and compound the returns, it adds up fast.

2Why does this even happen?

Prediction markets aren't like stock exchanges. There's no dedicated market maker quoting tight spreads on every pair. Polymarket uses a CLOB (Central Limit Order Book), which means prices are set by individual traders placing limit orders.

The lag creates the gap

When a big buyer sweeps the YES side, the YES price jumps. But the NO side doesn't automatically adjust. A different set of people have to come in and update their orders independently. That lag is where the arbitrage lives.

It's worse on lower-liquidity markets. On a presidential election market with $100M in volume, the spread is usually tight. But on a "Will X country ban Y by June?" market with $30K in volume, you'll regularly see 5-10% gaps.

Presidential Election ($100M+ volume)~0.5% gap

Mid-tier Market ($500K volume)~2-3% gap

Niche Market ($30K volume)5-10% gap

Single-market YES/NO arbs are the simplest case. The more interesting opportunities come from related markets that should be logically constrained but aren't.

Spot the Problem

Will Bitcoin hit $100K by June?$0.40

Will Bitcoin hit $150K by June?$0.25

If Bitcoin hits $150K, it must have also hit $100K. So the $100K market should be at least as high as the $150K market. The gap is only $0.15 when it should be much larger given the conditional relationship.

These logical constraints are everywhere on Polymarket:

Mutually Exclusive Events

If A and B can't both happen, their YES prices should sum to less than $1.00. When they don't, you sell both YES tokens.

Implication Chains

If A implies B, then B's price should always be ≥ A's price. When it's not, you buy B and sell A.

Exhaustive Sets

If a set of markets covers all possible outcomes, their prices should sum to $1.00. When they sum to more, you sell them all.

4Why it's harder than it sounds

If you're thinking "just write a script to scan for these," yeah, we thought so too. Here's where it gets tricky.

The four hidden pitfalls

Detecting related markets is hard

Polymarket doesn't tag markets as related. Figuring out that "Will Biden run in 2028?" and "Who will be the Democratic nominee in 2028?" are connected means parsing natural language across thousands of markets.

Execution risk is real

You find a 7% gap. By the time you get both legs filled, the gap might have closed. Your first trade can move the price enough to kill the edge on the second.

Resolution risk exists

What if the market gets voided? Or the resolution criteria are ambiguous? You could end up on the wrong side of a judgment call.

Capital lockup is the hidden cost

Most arbs resolve in days to months. Your capital earns 5-10% over that period. If a higher-yield strategy exists for the same timeframe, the arb may not be worth it.

5How we handle it

Our Two-Stage Filter Pipeline

Stage 1: LCMM Filter

Milliseconds

Fast mathematical pre-filter scans every market pair and flags anything where the numbers don't add up. Eliminates ~80% of markets instantly.

Stage 2: AI Validation

Seconds

Claude reads both market descriptions, checks logical relationships, and validates whether the apparent mispricing is real or if there's a hidden subtlety.

It's not foolproof. But it catches things that pure keyword matching would miss, and it's significantly cheaper than running AI on every single market pair from scratch.

The best arbs we've found tend to be in the 5-15% range on markets that resolve within 2-4 weeks. Anything larger than 15% is usually a trap.

5-15%Sweet spot edge

2-4 wkIdeal resolution

< 3%Not worth the risk

Anything larger than 15% is usually a trap: there's some resolution ambiguity or a reason the market is mispriced that isn't immediately obvious. And anything under 3% isn't worth the execution risk.