Sharp
Prices football and prediction markets with a calibrated model, then shows where the market is wrong. Fair odds, value edge, Kelly stake and a hedge, as a paid agent service (A2MCP) on OKX.AI.
视频
技术栈
描述
Sharp is a paid agent service (A2MCP) on OKX.AI that prices football and prediction markets. Most prediction agents guess a scoreline, or output probabilities that don't even add up to 100%. Sharp runs a Dixon-Coles model that returns calibrated win/draw/win probabilities that always sum to 100%, fair decimal odds, and the most likely scores.
It then compares that model price to the live market (a Polymarket query or supplied odds) and returns the value edge, expected value, a Kelly stake, and a hedge or arbitrage plan for each outcome. Every number is a deterministic function; an LLM never invents one.
Three services are exposed over MCP and REST: fair_odds (0.02 USDT), value_scan (0.10 USDT), and slip_builder (0.25 USDT). Any agent on OKX.AI can call it and pay per call via x402, settling in USDT0 on X Layer. Each result renders as a shareable card that is reproducible from its link. Registered on OKX.AI as Agent #6323, deployed on Vercel, open source under MIT.
本次黑客松进展
Built end to end during the hackathon: a Dixon-Coles bivariate Poisson model with strict probability normalization (fixing the common 125%-sum bug), a de-vig / value / Kelly / hedge engine as pure functions with 17 passing tests, three A2MCP services (fair_odds, value_scan, slip_builder), a spec-compliant x402 payment gate (HTTP 402 + PAYMENT-REQUIRED, USDT0 on X Layer chain 196), an MCP JSON-RPC endpoint, and stateless shareable report cards. Deployed to Vercel with GitHub auto-deploy, then registered and listed on OKX.AI as Agent #6323 in paid x402 mode.
融资状态
Not raising (bootstrapped, solo hackathon build).