Wash trading refers to the (illegal) practice of buying and selling securities without changing one’s net position, for the purpose of artificially inflating recorded volume. Evidence suggests that it is widespread on some exchanges where trader identities can be shielded. Wash trading can be hard to reliably distinguish from authentic and lawful strategies such as automated market making. We formulate wash trade detection as a graph clustering problem and apply the Leiden community detection algorithm to the network of all trades. The connectedness guarantees of the resulting partition imply a minimum economic cost that an adversary must incur in order to evade detection. Applying this method (and an alternate, iterative network-based procedure) to the Polymarket exchange, we estimate that wash trading peaked at nearly 60 percent of volume in December 2024 before subsiding substantially by May 2025, with continued activity until through summer 2026.