TypeScript CLI AI trading agent for Polymarket. It models World Cup match outcomes with a statistical (or optional LLM) engine, hunts for value versus live market prices, and can both gate copy-trades from whale / super-trader wallets and place its own autonomous value buys — plus manual Draw token control.

• AI match model — Elo + Poisson statistical engine, or an OpenAI-compatible LLM
  • Estimates Home / Draw / Away probabilities for each fixture
  • Blends model belief with vig-free market-implied probabilities
  • Finds edge, sizes positions with fractional Kelly, and reports expected value
  • Backtest harness with Brier score and log loss for calibration
• AI copy gate — only mirror a leader's trade when the model agrees there is edge
• Autonomous value scan — rank every open fixture by model edge and execute
• Copy trades from configured whale / super-trader wallets
• 4 copy strategies
  • scaled — mirror at 1 / copy_divisor
  • fixed_usd — fixed USD per copied trade
  • percent_leader — percentage of leader trade size
  • whale_gate — only copy when leader trade ≥ whale minimum, then scaled
• Interactive CLI: start/stop engine, settings, leader management, draw buys, AI menu
• World Cup market filter (optional)
• Dry-run mode for safe testing

cp .env.example .env
# Edit .env — set LEADER_WALLETS and keys for live trading

npm install
npm run dev

Adjust trade size, strategy, poll interval, draw defaults, World Cup filter, and the [ai] agent section (provider, edge / confidence thresholds, Kelly fraction, bankroll, market blend weight, temperature) without code changes.

1. Start copy trading — polls leader activity and mirrors new trades (AI-gated when enabled)
2. Stop copy trading
3. Settings — amount, strategy, min order, whale gate, dry-run
4. Buy Draw token — pick a FIFA 2026 match and buy the Draw outcome manually
5. Manage leaders — add wallets or import from Polymarket leaderboard
6. AI agent — analyze a match, scan fixtures for value, backtest, AI settings
7. Status
8. Quit

The agent turns ratings + Elo into expected goals, runs a Poisson model for Home / Draw / Away probabilities, blends that with the vig-free market, and only acts when both the edge and confidence clear the configured thresholds.

• Analyze a match — probabilities, per-outcome edge, EV, and a sized signal
• Scan fixtures for value — ranks every open fixture by edge and can execute
• Copy gate — vetoes mirrored buys the model disagrees with (fails open on errors)
• Backtest — replays the model over sample fixtures and reports ROI, Brier, log loss

Set AI_PROVIDER=llm (with AI_API_KEY) to back the agent with any OpenAI-compatible endpoint; otherwise the offline statistical model is used.

LEADER_WALLETS=whale:0x204f72f35326db932158cba6adff0b9a1da95e14:WhaleAlpha,trader:0x56687bf447db6ffa42ffe2204a05edaa20f55839:TopTrader

Prediction market trading involves risk. Test with DRY_RUN=true first. Ensure you comply with Polymarket geographic restrictions and local laws.