oracle3

Oracle3

Oracle3 is an open-source trading engine and MCP server for prediction markets. It maps the logical relations between event contracts, finds prices that break the axioms of probability after each venue's fees, and trades them live on Kalshi, Polymarket and Solana, or on paper, under pre-trade risk limits.

Trades live on Kalshi, Polymarket and Solana. oracle3 live run executes with the same engine that runs paper trading, behind pre-trade risk limits and a kill switch. AI agents plug in through a 13-tool MCP server.

At a glance

Venues	Kalshi, Polymarket and Solana (DFlow)
Execution	Live and paper on one engine, with pre-trade risk limits, a kill switch and Jito bundle submission on Solana
Relations checked	implication, exclusivity, complement, same event across venues, event sum
Costs	Each market's own fee schedule from the venue API (Kalshi taker 0.07·M·C·P·(1−P); Polymarket taker rate·C·p·(1−p))
Strategies	6 constraint-based, 2 statistical-arbitrage, 2 model-driven
Agent interfaces	MCP server with 13 tools, JSON CLI, 6 agent skills, Python API
Tests	600+, with ruff, mypy and codespell in CI
Install	pip install oracle3
License	Apache-2.0; the original U Lab portions are MIT (see NOTICE)

What problem does it solve?

Contracts on related outcomes are tied together by probability. If A implies B, then P(A) ≤ P(B). If A and B cannot both happen, P(A) + P(B) ≤ 1. The outcomes of one event sum to one. Quoted prices break these bounds, within a venue and across venues, and a basket of contracts that pays a known amount in every state can then be bought for less than that amount.

The gaps are small, and both venues charge taker fees that scale with p(1 − p). Whether a gap is worth anything depends on the fee on every leg of the basket. Oracle3 does three things with that:

1. Relations. It records which markets are related and how (implication, exclusivity, complement, same event, event sum).
2. Checks. For each relation it finds the cheapest basket at executable prices, prices every leg under that market's own fee schedule, and reports the edge before and after fees.
3. Execution. It trades the baskets that survive, live or on paper, under position, drawdown and exposure limits, with a kill switch.

How do I run it?

pip install oracle3

# Find markets (JSON output for scripts and agents)
oracle3 market search --exchange kalshi --query "fed" --json
oracle3 market search --exchange polymarket --query "fed decision" --json

# Start the MCP server over stdio
oracle3 mcp

# Trade live on Kalshi (key via KALSHI_API_KEY_ID and KALSHI_PRIVATE_KEY_PATH)
oracle3 live run --exchange kalshi --monitor \
  --strategy-ref oracle3.strategy.contrib.implication_arb_strategy:ImplicationArbStrategy

From Python:

from oracle3.arbitrage import Quote, check_constraint
from oracle3.fees import KalshiSchedule

# A implies B, but A is bid at 0.60 while B is offered at 0.55.
result = check_constraint(
    "implication",
    [Quote("A", yes_bid=0.60, schedule=KalshiSchedule()),
     Quote("B", yes_ask=0.55, schedule=KalshiSchedule())],
)
best = result.best
print(best.description, best.gross_edge, best.fees, best.net_edge)
# NO on A + YES on B 0.05 0.0342 0.0158

Full CLI reference: documentation.

How do AI agents use it?

MCP server

Add it to any MCP client. For Claude Code:

claude mcp add oracle3 -- uvx oracle3 mcp

For Claude Desktop, Cursor and other clients that read an mcpServers block:

{
  "mcpServers": {
    "oracle3": { "command": "uvx", "args": ["oracle3", "mcp"] }
  }
}

Tool	What it does	Side effects
search_markets	Keyword search on Kalshi or Polymarket; Kalshi series listing	read-only
get_market	Prices, volume, close time and resolution rules	read-only
get_orderbook	Both sides of the book, best level first	read-only
get_quote	Best bid and ask on YES and NO, with the market's fee schedule	read-only
check_constraint_live	Fetch quotes and fee schedules, then check a relation	read-only
check_constraint	Check a relation on quotes you supply	none
trading_fee	Fee for one fill under a venue schedule	none
fair_value	Probability implied by a price under the Wang transform	none
list_relation_types	The supported relations and their bounds	none
list_relations	Relations saved locally by the research CLI	reads a local file
paper_order	Buy in a local paper ledger, filling against the live book with fees	writes a local file
paper_portfolio	Cash, positions and fills in the paper ledger	reads a local file
paper_reset	Erase the paper ledger (requires confirm=true)	writes a local file

Real-money execution stays in the CLI: agents research and paper-trade through MCP, and a human signs off on live orders.

If your client ran oracle3 1.2.0, which failed to start with mcp 2.x, refresh uv's cached copy once with uvx --refresh oracle3 mcp.

Agent skills

skills/ (mirrored in .claude/skills/ for Claude Code) holds step-by-step instructions for agents:

Skill	Use it to
pm-constraint-arbitrage	Check related markets for a fee-surviving violation with the MCP tools
pm-data-discovery	Find markets and save research samples
pm-quant-strategy-authoring	Write a tunable QuantStrategy
pm-agent-strategy-authoring	Write an LLM- or tool-driven AgentStrategy
pm-paper-trade-ops	Run, monitor and archive paper trading
pm-live-trade-ops	Live trading, only with explicit user approval

JSON CLI

Every market, paper and trade command, and every research command except research memory, accepts --json. A running engine can be paused, resumed, inspected and stopped from another process with oracle3 trade pause|resume|state|stop --json. See AGENTS.md for which commands are read-only.

What do fees do to the edge?

Both venues charge taker fees proportional to p(1 − p). A two-leg taker basket with both legs near 0.50 has to clear these violations per contract before any edge is left:

Venues	Break-even violation
Kalshi + Kalshi	3.50¢
Kalshi + Polymarket (rate 0.05)	3.00¢
Kalshi + Polymarket (rate 0.04)	2.75¢
Polymarket + Polymarket (rate 0.04)	2.00¢

Buying every outcome of an n-way event costs k(1 − Σp²) per contract, which approaches 7¢ on Kalshi as outcomes multiply. The derivation, the tables and the sources are in Do prediction-market arbitrage edges survive fees?; python scripts/fee_frontier.py reproduces every number.

How is it tested?

• oracle3.fees reproduces Kalshi's published fee table and Polymarket's documented fee example.
• oracle3.arbitrage is unit-tested for every relation, including mixed-venue baskets and missing quotes.
• The MCP server is tested against mocked venue APIs, run against the live public APIs, and checked in CI on both major versions of the MCP SDK.
• The pricing engine uses the coefficients from the companion working paper (SSRN 6468338), checked against its replication package.

Roadmap

1. Price every strategy signal with the venue fee schedules in oracle3.fees.
2. Measure how often and how deeply live violations clear the fee hurdle, per relation and venue pair.
3. Wire SpreadExecutor, multi-leg execution with LIFO unwind on partial fills, into the multi-leg strategies.
4. Publish a pre-registered forward track record with timestamped daily snapshots.

How is it built?

graph TD
    R[Relation store<br/>implication · exclusivity · complement · same event · event sum] --> C[Constraint checker<br/>oracle3.arbitrage + oracle3.fees]
    Q[Venue data<br/>Kalshi · Polymarket public APIs] --> C
    C --> S[Strategy layer<br/>6 constraint-based · 2 statistical · 2 model-driven · LLM agents]
    P[Pricing engine<br/>Wang transform, calibrated in Yang 2026] --> S
    S --> E[Trading engine<br/>risk manager · position tracker · kill switch]
    E --> T[Paper trader]
    E --> L[Live traders<br/>CLI only]
    C --> M[MCP server<br/>read-only tools + paper ledger]
    Q --> M

Relations and venue quotes feed the constraint checker, which prices every basket under each market's fee schedule. Strategies consume those checks and the pricing engine's fair values and send orders through a trading engine that enforces risk limits. The MCP server exposes the data, the checker and a separate paper ledger to agents; live traders are reachable only from the CLI.

Constraint-based strategies, each enforcing one probability bound:

Strategy	Bound
Cross-market	Same event, same price across venues
Exclusivity	P(A) + P(B) ≤ 1 for mutually exclusive events
Implication	P(A) ≤ P(B) when A implies B
Conditional	P(A | B) within derived bounds
Event sum	Σ P(outcome) = 1 within an event
Structural	P(A) = β·P(B) + α from a fitted relation

Statistical arbitrage: cointegration spread, lead-lag. Model-driven: fair-value divergence and premium decay, using the pricing model below.

Pricing model. Fair values come from the Wang transform p_mkt = Φ(Φ⁻¹(p) + λ), with λ estimated on 291,309 resolved contracts in the companion working paper, Yang (2026), Pricing Prediction Markets: Incomplete Markets, Selection Rules, and Calibration Wedges (SSRN 6468338). The model and its estimates are documented there.

Related projects

• ulab-uiuc/prediction-market-cli (Coinjure): agent-native trading system for prediction markets; Oracle3 bundles its coinjure package.
• pmxt-dev/pmxt: unified API across prediction-market venues.
• Jon-Becker/prediction-market-analysis: data collection and analysis framework with a large public dataset.
• YichengYang-Ethan/prediction-market-pricing: replication package for the pricing model.

How can I collaborate?

• Open problems are tracked as issues labeled open-problem: measuring violations against the fee hurdle, evaluating relation discovery, and comparing LLM and market calibration.
• Discussions are open for questions and ideas: GitHub Discussions.
• Contributions: see CONTRIBUTING.md.

How do I cite it?

Citation metadata is in CITATION.cff, and every release is archived on Zenodo (DOI 10.5281/zenodo.20062548). For the pricing model, cite the working paper (SSRN 6468338).

Origin and attribution

Oracle3 began as ulab-uiuc/oracle3, developed by Yicheng Yang and Haofei Yu at U Lab (University of Illinois Urbana-Champaign) under the MIT License, and it bundles the coinjure package from the same lab. The strategy, pricing, risk, dashboard, and test layers in this repository were added on top of that base; see NOTICE for the retained license text.

License

Apache 2.0; see LICENSE. Portions from the original U Lab code remain under the MIT License reproduced in NOTICE.

This software is for research and education. Trading involves financial risk.