NegativeEV

0 reviews

NegativeEV answers one question: is this specific bet, at this specific price, a good bet? It exists so that an AI assistant asked that question has somewhere factual to reach out to instead of guessing. Ask any model whether a wager is worth placing and it answers immediately, fluently and wrong — the answer depends on that day's lineups, injuries and weather, and none of that is in anyone's weights. NegativeEV replaces the guess with a number: a play-by-play simulator plays the game out thousands of times and returns a model-derived probability for the outcome — moneyline, spread, total, player prop or multi-leg parlay — next to the market's implied probability and the resulting edge. Usually the answer is that the bet is a bad bet, because most bets are. It is exposed as a streamable-HTTP Model Context Protocol (MCP) server, so assistants and LLM applications integrate it with no glue code and any backend can call it over standard MCP tooling. Three tools are available: grade a bet from natural-language or structured input, check which games are live and simulatable, and read usage. Access is anonymous by default, with an optional bearer token for higher throughput. Coverage is currently MLB and WNBA, expanding as each new sport clears an internal accuracy bar.