yichengyang-ethan/oracle3-prediction-market-agent/pm-constraint-arbitrage

PM Constraint Arbitrage

Use this skill when the user asks whether related markets are mispriced relative to each other, for example "does P(A) exceed P(B) even though A implies B?", or whether a cross-venue price gap survives fees.

Tools

MCP server (pip install oracle3, then run oracle3-mcp):

• search_markets, get_quote, get_orderbook to find markets and read executable prices.
• check_constraint_live to fetch quotes and fee schedules and evaluate a relation.
• check_constraint to evaluate a relation on quotes you supply.
• trading_fee to price a single fill under a venue's schedule.
• paper_order, paper_portfolio to paper-trade the legs.

Python API: oracle3.arbitrage.check_constraint and oracle3.fees.

Relations

Steps

1. Find the markets with search_markets and read both markets' resolution rules with get_market.
2. Decide which relation holds. This is a judgment about the contract wording, not something the tool can verify.
3. Run check_constraint_live with the relation and the markets in the order above.
4. Read best.gross_edge_per_contract (edge before fees), best.fees_per_contract and best.net_edge_per_contract.
5. Only if the net edge is positive and the book is deep enough, paper-trade each leg with paper_order at a limit no worse than the quoted ask.

Hard rules

• Never place real orders from this skill; the MCP server cannot, and the CLI live mode is out of scope.
• State the relation you assumed and why; a wrong relation turns an "arbitrage" into a directional bet.
• Report fees per leg. Kalshi taker fees are 0.07·M·C·P·(1−P); Polymarket taker fees are rate·C·p·(1−p) with a per-market rate; Polymarket makers pay nothing.
• The check assumes every leg fills at the quoted price. Say so when the displayed size is smaller than the order.