Live Sports Score Arbitrage

Real-time live score arbitrage against Polymarket sports markets using ESPN data.

What It Does

Every 2 minutes, pulls live scores from ESPN's public API and calculates fair win probabilities using sport-specific statistical models:

| Sport | Model | Key Parameter | |-------|-------|--------------| | NBA | Gaussian | sigma = 12.5 points | | NFL | Gaussian | sigma = 13.5 points | | NHL | Gaussian | sigma = 1.5 goals | | Soccer | Poisson | lambda = 1.35 goals/90min |

Gaussian model: P(win) = CDF(lead / (sigma * sqrt(t_remaining / T)))

Poisson model: Calculates expected goals from current rate, sums probability of all winning scorelines.

Edge Source

Polymarket prices update slowly during live games - often 30-120 seconds behind real scoring events. ESPN's API updates near-instantly. This latency gap is the edge.

Scheduling

Runs every 2 minutes via cron (*/2 * * * *). Managed automaton (auto-executes on schedule). Only trades when divergence exceeds 10%.

Requirements

pip dependencies: simmer-sdk, requests

Environment variables:

• SIMMER_API_KEY (required) - get from simmer.markets/dashboard
• TRADING_VENUE (optional) - defaults to sim for paper trading, set to polymarket for real
• SPORTS_DIVERGENCE_THRESHOLD (optional) - minimum divergence to trigger trade, defaults to 0.10 (10%)
• SPORTS_TRADE_SIZE (optional) - trade size in USD, defaults to 20.0

Usage

python sports_live.py                  # dry run (default, no trades)
python sports_live.py --live           # real trades via SimmerClient
python sports_live.py --live --quiet   # cron mode

Dry-run by default. Pass --live to execute real trades.

Trade Execution Path

1. sports_live.py fetches live scores from ESPN public API (no auth needed)
2. Calculates fair win probability using Gaussian (NBA/NFL/NHL) or Poisson (soccer) models
3. Finds matching Polymarket markets via SimmerClient.find_markets()
4. Compares model probability to Polymarket price, applies >10% divergence threshold
5. If --live: calls SimmerClient.trade() with market_id, side, amount, reasoning
6. If dry-run (default): logs the signal but does NOT execute any trade

Security

• All trades go through SimmerClient.trade() only. No direct CLOB or wallet access.
• Reads SIMMER_API_KEY and optionally TRADING_VENUE from env. No other secrets.
• No wallet private keys are required or read by this script.
• Dry-run is the default. The --live flag must be explicitly passed to execute trades.

Remixable Template: Fork this skill to adjust sigma values, add new sports (MLB, tennis), change the divergence threshold, or plug in different live score sources.