Polymarket Scanner
Finding prediction markets that don't add up
- Role
- Personal project
- Year
- 2026
- Status
- Shipped
The problem
A read-only scanner that sweeps liquid Polymarket markets looking for two specific mispricings: binary pairs whose complementary asks sum to less than a dollar, and multi-outcome groups whose outcomes collectively price below certainty.
In a well-formed prediction market the Yes and No sides of a binary question should cost about a dollar together, and the outcomes of a multi-outcome event should sum to about a dollar too. Sometimes they don't — thin books, stale quotes, a market nobody has looked at in an hour. Those windows are small, and finding them by hand across a thousand markets is not something a person can do.
How I built it
Read-only on purpose
It never places an order. It prints candidates and stops. Automated execution against a live book is a different project with a different risk profile, and conflating the two is how research tools quietly become trading bots that lose money while you're asleep. The scanner's job ends at 'this looks wrong, go and check'.
Two patterns, one of them the interesting one
The binary case is easy to state: best ask on Yes plus best ask on No coming in under a dollar. The multi-outcome negRisk case is the one worth having — buying a group of outcomes that together price below certainty. It's the same pattern the largest participants use, at a scale where the arithmetic works out.
Priced against reality, not the screen
A candidate is only real after fees, slippage against actual book depth, and the capital locked up until resolution. The scanner reports against depth rather than the top-of-book quote, because most apparent edge evaporates the moment you ask how much you could actually fill.
Built with
- Python
- Polymarket CLOB API
- asyncio
Outcome
- Sweeps roughly a thousand liquid binary markets per run
- Detects both complementary-token and multi-outcome negRisk mispricings
- Read-only by design — surfaces candidates, never executes