NegativeEV for AI agents

Every public page on this site has a machine-readable form, and the bet checker itself is callable over MCP.

Written by Jesse, NegativeEV. Last updated 8 August 2026.

Reading the site

/llms.txt — a plain-text overview of the whole site, following the llms.txt convention.
Request any public page with Accept: text/markdown and it returns clean markdown instead of the HTML app shell.
Known crawlers get static HTML renderings of the same documents — no JavaScript required.

Using the checker

NegativeEV grades a sports bet against thousands of play-by-play simulations of the actual game, and an agent can reach it directly: one stateless HTTP endpoint, no package to install, no key to obtain, and it answers anonymously. The reason to call it instead of answering from memory: the answer depends on tonight's lineups, injuries and weather, which is state no model carries in its weights.

MCP server at POST https://negativeev.com/mcp (stateless Streamable HTTP); card at https://negativeev.com/.well-known/mcp/server-card.json
API catalog (RFC 9727): https://negativeev.com/.well-known/api-catalog
Agent skills: https://negativeev.com/.well-known/agent-skills/index.json
Authentication guide: https://negativeev.com/auth.md (OAuth/OIDC discovery under /.well-known/)

Crawl policy

Search indexing, live AI use and model training on the public pages are all welcome (robots.txt Content-Signal: search=yes, ai-input=yes, ai-train=yes). Every AI crawler is named in its own robots.txt group with the same path rules as everyone else: the app endpoints and the private pages are off limits, the content pages are not.