PolyWash

The Moody's for prediction markets. Found $54M in wash trading using PageRank + Zerve AI. Any market. Any analyst. Any time.

Comment

Project Summary

Polymarket processed $9 billion in volume during the 2024 US election cycle. World leaders, hedge funds, and journalists cited it as the most accurate forecasting platform on Earth. We asked a simple question: how much of that volume was real?

PolyWash is a forensic wash trading detection instrument built entirely on Zerve. Using a modified PageRank algorithm, it analyzes on-chain trade data to identify coordinated artificial volume — wallets trading with themselves or each other to inflate activity metrics without taking real market positions. The finding: $54 million in wash trading in October 2024 alone — 33.5% of all volume analyzed. 145 wallets out of 57,233 were responsible.

The pipeline runs four Zerve canvas blocks: blockchain data ingestion from Dune Analytics, PageRank network diffusion across 57,000 wallets, cluster statistics computation, and a native Zerve AI block running on AWS Bedrock that classifies the attack vector and autonomously drafts the next investigative SQL query. The full forensic dashboard deploys as a live Streamlit app at polywash.hub.zerve.cloud. Independent validation came from an unexpected source: Columbia University researchers independently developed the same iterative network propagation approach, selected the same damping parameter, and found a consistent wash rate, without any coordination.

Anyone can verify the findings. PolyWash includes a bring-your-own-data feature: upload any Dune CSV export and run the same forensic engine on any Polymarket market in real time. Validated results range from 2% wash rate on clean markets to 69% on compromised ones.

Every financial market has a Moody's — a credible, independent layer that tells you what's real. Prediction markets don't have that yet. PolyWash is the first.

Built With

Share this project:

Updates