README

Prove is a statistical instrument for finding where autonomous and agentic systems break. It measures failure rather than simulating, training, or benchmarking, and it stays out of the environment and the user’s code.

Serve

serve holds a read-only page for one stored pair on an ephemeral loopback port: the plain summary first, then the statistics it came from.

poisson-prove --store ./prove-runs serve A B --event NAME --x AXIS

It binds nothing until the page is built, fetches nothing, and writes nothing; --json prints the page’s structure instead. What each section answers is in Reading a serve page.

Docs

• Reading a serve page — what each of the fourteen sections answers, and what it cannot
• Defining outcomes — an outcome names what happened to the system under test, never what happened to the instrument. The rule, the case that taught it, and a checklist to run before the first episode.
• What capture stores — what content = false strips, what it keeps on purpose, and how max_observation_bytes bounds one observation value.