Adam v2 is an autonomous research and trading system for the Polymarket prediction market. It scans the entire market, models fair value for every question, shadow-tracks every opportunity, validates each strategy on paper, and deploys real money only on strategies that have proven a positive expected value. It is a personal system with no public URL; this page is an engineering case study.
Why it exists
Research shows roughly 92% of Polymarket traders lose money. The first version of this bot lost too, for structural reasons: buying favorites with a naive price-gap heuristic, crossing the spread on every order, and flipping strategy after just two bets. Version 2 fixes this structurally: paper-first by default, shadow-tracking of every candidate signal (including bets it does not take), real edge gates priced at the order-book mid, and graduation to live trading only after enough resolved paper bets show a confidence-bounded positive EV. Even then, real money moves only when the operator manually flips the switch.
The four strategies
- Structural arbitrage: detects and executes mathematical price inconsistencies, verifying the complete outcome set and pricing every leg at real order-book depth
- Copytrade: auto-discovers top traders from the Polymarket leaderboard by ROI and PnL, mirroring their fresh buys
- Crypto: a volatility model computing true probabilities from Binance data
- Live sports: comparing ESPN live win probabilities to market prices
Engineering
Python throughout: an engine loop (resolve, fetch, strategize, shadow-log, gate, trade, report), a single persistence layer on SQLite or Postgres (Railway), fractional Kelly sizing with daily loss limits and exposure caps, a FastAPI dark dashboard with a strategy scoreboard and model calibration chart, and a daily Telegram digest. Access is private. No bot can guarantee profit; this one guarantees it will not risk money before an edge is proven on data.