Forezai runs 24 paper-trading experiments side-by-side against Polymarket's 5-minute BTC, ETH, SOL and XRP Up/Down markets — sniper, fair-value taker, maker-quoter, and regime-switch strategies, with live calibration analysis. It's open source, MIT licensed, and built to tell you when your strategy doesn't work.
Forezai is an open-source harness for stress-testing prediction-market trading strategies against real, live order books — without putting real money on the line. The dashboard shows you, in real time, exactly how much each strategy is making or losing, why each decision was taken, and how close every experiment is to its risk gates.
Most of the strategies in Forezai's fleet do not make money. An honest analysis of the first ~700 paper trades found exactly one strategy with the statistical signature of real edge — and even that one's sample is still too small to confirm. Win rate alone is misleading: a 90% win rate at $0.95 entries is structurally below breakeven.
This project is shared so other researchers can run the same experiments, see the same results, and reach the same uncomfortable conclusion: prediction-market trading is harder than it looks, and most "edges" are mechanical illusions. Forezai is the instrument; the data is the answer.
If you are looking for a money-printing bot, this is not it. If you are looking for an honest research platform to investigate where edge might live, you may find it useful.
A Node.js bot reads live data from Polymarket's CLOB + Chainlink + Binance, runs each experiment's strategy independently with its own paper bankroll, and streams the full state to a Next.js dashboard over WebSocket at 4 Hz. Every fill, every settle, every risk-gate check is written to an append-only audit log.
The bot connects to Polymarket Gamma for market discovery, Polymarket CLOB for live order books, Polymarket RTDS (or Pyth / Chainlink Direct) for the resolution-grade price feed, and Binance for the chart-grade tape.
Each experiment's Engine instance ticks 4× / second. Per tick: classify the current vol regime, evaluate the configured strategy (sniper / fair-value / maker / regime-switch), and log every decision — including the skipped ones, with reasons.
In paper mode the fill walks the real book at the real ask, applies the real Polymarket fee schedule (0.07 × p × (1 − p)), and adds a simulated 2.5-second latency to mirror RPC + sign + submit time. Window misses count as misses.
Markets resolve every 5 minutes on the Chainlink BTC/USD reading. Wins pay $1.00 / share; losses pay $0. Realized P&L hits the day's running counter; the consecutive-loss + daily-loss + drawdown gates check on every settle.
The bot pushes a full state object to the dashboard over WebSocket every 250 ms: every experiment's equity, P&L, fills, positions, current decisions, calibration buckets. One connection regardless of how many experiments are running.
The dashboard's Fleet view ranks all 24 experiments by P&L with risk-headroom bars on every row. Click a row → a Focus drawer opens with Live / Trades / Analytics / Strategy / Decisions sub-tabs for that one experiment.
Risk gates can be tuned per-experiment from the UI. Bankrolls can be reset to any custom amount. Fleet presets save the current configuration to localStorage so a research session can be restored later.
Every experiment in the fleet runs one of these four strategy modes. The same Engine code, the same risk gates, the same paper-fill mechanics — only the firing logic differs.
Late in the window, buy the side the market already favours (e.g. $0.90 ≤ ask ≤ $0.96). High win rate by construction — but breakeven sits at the favourite's implied probability, so 90% wins at $0.95 is structurally negative.
Brownian-motion model of the underlying. Buy whichever side the model thinks is mispriced by more than edgeMargin. Lower win rate, asymmetric payouts when right. The only strategy in the fleet showing the math signature of real edge so far.
Posts symmetric quotes a few cents below fair on both sides. Earns the spread when traders cross — no taker fees. Direction-agnostic. The newest strategy in the fleet; data is still thin.
Classifies current 5-min volatility into low / mid / high, then routes to a sub-strategy. Low → sniper, mid → fair-value, high → maker. Adaptive in theory; in practice, switching adds friction that subtracts more value than it captures.
Forezai is published on GitHub, MIT-licensed, with 104 passing unit tests, a full CHANGELOG of every PR, and explicit honest disclaimers in the README. Fork it, run your own experiments, propose your own strategies, share your findings. The only requirement is that you preserve the MIT attribution and the risk warnings.
Thorsten Meyer is a Munich-based futurist, post-labor economist, and recipient of OpenAI's 10 Billion Token Award. He spent two decades managing €1B+ portfolios in enterprise ICT before deciding that writing about the transition was more useful than managing quarterly slides through it. More at ThorstenMeyerAI.com.
Forezai is one of several open research projects he publishes. Built independently with the assistance of large-language-model coding tools, then carefully reviewed and shipped under the MIT license. The project's bias is honesty over hype: when the data says a strategy is failing, the dashboard says so. When a high win rate is just a mechanical illusion, the calibration view explains why.
Companion projects: ThorstenMeyerAI.com (essays on the post-labor transition) and DojoClaw (the AI content engine behind 450+ magazines).
Forezai's LIVE path can lose every cent of any wallet you connect to it. Paper-trading P&L is not predictive of real-money outcomes. Polymarket fees, adverse selection, RPC failures, network latency, and strategy bugs each — independently — can drain a funded wallet in minutes. The software is provided as is with no warranty. If you are not prepared to lose 100% of any funds you commit, do not flip LIVE_ENABLED=true. This is research, not investment advice.