Polymarket quant bots revealed
A public deep dive into Polymarket trading shows quant bots using PhD‑level methods—Bayes, Kelly Criterion and sub‑40ms signal‑to‑order latency—and a disclosed run returned about $250K in two weeks. It’s a blunt reminder that HFT-style edge combines advanced math with engineering and ops discipline. (x.com)
0x_Discover expanded the X thread into a longer technical post and a Telegram thread that names specific stack components—MiroFish for swarm-style market sims, OpenClaw for always-on execution, and Claude Opus for LLM-driven strategy reasoning. (threadreaderapp.com) An independent, published buildout shows an end‑to‑end ML pipeline that pulls Polymarket CLOB/Gamma data, trains a five‑model stacking ensemble (XGBoost, LightGBM, HistGradientBoosting, ExtraTrees, RandomForest) and applies probability calibration plus fractional position sizing. (navnoorbawa.substack.com) Several public GitHub projects demonstrate LMSR‑aware signal processors and posterior‑update style estimators wired into Polymarket order routers, with accompanying position‑sizing modules and automated stop/circuit rules. (github.com) The deeper posts and accompanying repos document execution pipelines that push latency into the millisecond regime, and multiple infrastructure projects publish benchmarks and architecture notes showing tens‑of‑milliseconds signal‑to‑order timings for competitive deployments. (github.com) Performance claims linked to the thread include a Telegram post asserting roughly $400K in gains in under a month from an automated OpenClaw setup and a separate viral claim of turning a $12K deployment into about $43.8K within hours by exploiting timezone and liquidity windows. (t.me) Public tooling for replication and analysis has proliferated: real‑time bot leaderboards and replay platforms (Polybot Arena), sub‑penny market scanners (PolySnipes), and a Polymarket backtesting API (PolyTest) now coexist with community bot repos that include full backtest and monitoring stacks. (polybot-arena.com)
