Polymarket Quant Trader

## Iteration 4 (Auto 4/8)
- Changed: minEdgePct 2 → 3
- Brier: 0.1804 (prev: 0.1814)
- Improvement: +0.0010
- Status: ✅ KEPT — new best
- Version: 1.0.1

Polymarket YES price: $0.40 (implied 40%)
1WIN decimal odds:    2.80  (implied 1/2.80 = 35.7%)

Spread = |40% - 35.7%| = 4.3%

// spread-calculator.ts
export function calcSpread(polyProb: number, onewinDecimalOdds: number) {
  const onewinProb = 1 / onewinDecimalOdds;
  const spread = Math.abs(polyProb - onewinProb);
  const spreadPct = spread * 100;
  const direction = polyProb < onewinProb ? "buy_poly_yes" : "buy_poly_no";
  return { onewinProb, spread, spreadPct, direction };
}

export function calcExpectedProfit(polyProb: number, onewinProb: number): number {
  const edge = Math.abs(polyProb - onewinProb);
  const ONEWIN_VIG = 0.02;
  return Math.max((edge - ONEWIN_VIG) * 100, 0);
}

export function calcKellyFraction(polyProb: number, onewinProb: number): number {
  const edge = Math.abs(polyProb - onewinProb);
  const fraction = edge / (1 - Math.min(polyProb, onewinProb));
  return Math.min(fraction, 0.10); // Hard cap at 10%
}

export function getConfidence(spreadPct: number): "HIGH"|"MEDIUM"|"LOW"|null {
  if (spreadPct > 5) return "HIGH";
  if (spreadPct >= 3) return "MEDIUM";
  if (spreadPct >= 1) return "LOW";
  return null; // Skip
}

npm run arb:scan

🟢 HIGH CONFIDENCE | Spread: 6.2%
  PM: "Will Bitcoin hit $100k by March?" @ $0.35
  1WIN: Same event @ 2.50 odds (40.0%)
  Direction: buy_poly_yes
  Kelly: 4.8% of bankroll
  Expected profit: 4.2%

🟡 MEDIUM CONFIDENCE | Spread: 3.8%
  PM: "Lakers vs Celtics Game 5 winner" @ $0.55
  1WIN: Same event @ 1.72 odds (58.1%)
  Direction: buy_poly_no
  Kelly: 2.1% of bankroll
  Expected profit: 1.8%

// title-matcher.ts
// Normalizes: lowercase, remove punctuation, strip stop words
// Stop words: vs, v, the, will, who, win, to, in, at, on, a, an,
//   of, for, and, or, be, is, are, was, match, game, fight, bout

// Dice coefficient: 2 * |intersection| / (|a| + |b|)
// Threshold: 0.4 minimum for a match

// detector.ts — startMonitor()
// Polls every 60 seconds
// Tracks seen arb IDs to alert only on NEW opportunities
// Logs all discoveries with timestamps

git clone <your-polymarket-bot-repo>
cd polymarket-bot
npm install

# Required for live trading
POLYGON_WALLET_PRIVATE_KEY=your_polygon_private_key
POLYMARKET_FUNDER_ADDRESS=your_funder_address
POLYMARKET_API_URL=https://gamma-api.polymarket.com
POLYMARKET_CLOB_URL=https://clob.polymarket.com

# Risk management
STARTING_CAPITAL=1000
MAX_POSITION_SIZE=500
MAX_TOTAL_EXPOSURE=2000
MIN_EDGE_THRESHOLD=0.10
STOP_LOSS_PERCENT=5
TAKE_PROFIT_PERCENT=5

# Optional: CEX for hedging
BINANCE_API_KEY=
BINANCE_API_SECRET=

# Safety
DRY_RUN=true        # Start with paper trading!
LOG_LEVEL=info
LOG_TO_FILE=true

# Stream 1: Score markets and find EV opportunities
npm run agent:alpha

# Stream 2: Run strategy optimizer overnight
npm run research:auto

# Stream 3: Scan for cross-platform arb
npm run arb:scan

# Full bot (all streams)
npm run bot

# 1. Scan for spreads
npm run arb:scan

# 2. Review HIGH confidence opportunities only
# Look for spreadPct > 5% with Kelly > 3%

# 3. Verify the match manually
# Check that the title matcher correctly paired the events
# Open both Polymarket and 1WIN to confirm prices are live

# 4. If confirmed, execute on Polymarket (DRY_RUN=false)
# The bot respects MAX_POSITION_SIZE and STOP_LOSS_PERCENT

# 1. Check current Brier score
npm run research:eval

# 2. Start the auto-improver (takes 10-30 minutes)
npm run research:auto

# 3. Check results in the morning
cat research/program.md | tail -30

# 4. If version bumped, verify the checkpoint
ls research/checkpoints/

# 5. Deploy updated strategy
# The agent automatically uses the latest strategy.ts

// Run in your TypeScript environment
import { scoreMarket, rankByEV } from './src/quant/ev-calculator';
import { quarterKelly } from './src/quant/kelly-criterion';

// For each active position, score against your current probability
const positions = [
  { id: 'btc-100k', yesPrice: 0.35, ourP: 0.42 },
  { id: 'election-x', yesPrice: 0.60, ourP: 0.55 },
];

const scored = positions.map(p => scoreMarket(p, p.ourP));
const ranked = rankByEV(scored);

ranked.forEach(m => {
  console.log(`${m.marketId}: EV=${m.ev.toFixed(3)}, Edge=${m.edgePct.toFixed(1)}%, Kelly=${m.kellyFraction.toFixed(3)}`);
  console.log(`  → ${m.recommend ? '✅ TRADE' : '⏭️ SKIP'}`);
});

import { bayesUpdate, addEvidence } from './src/quant/bayesian-updater';

// Initialize state for a new market
let state = {
  marketId: 'fed-rate-cut-march',
  priorP: 0.50,
  currentP: 0.50,
  evidence: [],
  lastUpdated: new Date(),
};

// New evidence: Fed minutes suggest dovish stance (LR > 1 → supports YES)
state = addEvidence(state, {
  description: 'Fed minutes dovish tone, multiple members favor cut',
  likelihoodRatio: 1.8,
  timestamp: new Date(),
});

console.log(`Updated probability: ${(state.currentP * 100).toFixed(1)}%`);
// Output: ~64.3% (moved from 50% toward YES)

// More evidence: CPI comes in hot (LR < 1 → supports NO)
state = addEvidence(state, {
  description: 'CPI +0.4% MoM, above expectations',
  likelihoodRatio: 0.6,
  timestamp: new Date(),
});

console.log(`Updated probability: ${(state.currentP * 100).toFixed(1)}%`);
// Pulled back toward 50%

// 1. Edit research/strategy.ts with your hypothesis
// Example: change minEdgePct from 3.0 to 2.0

// 2. Run evaluation
// npm run research:eval

// 3. Check the backtest output:
// BacktestResult {
//   winRate: 0.54,
//   brierScore: 0.1820,
//   sharpeEstimate: 1.2,
//   recommendation: 'EDGE'  // or 'STRONG_EDGE' / 'NO_EDGE'
// }

// Decision matrix:
// STRONG_EDGE (winRate > 56% AND Brier < 0.22) → Deploy immediately
// EDGE (winRate > 52%) → Run for 1 week on paper
// NO_EDGE → Revert the change