The words for trusting AI-written code.

01Why a glossary

Trust needs a shared vocabulary.

Teams adopting coding agents keep running into the same questions with different words for them. Are the tests any good? Did the change do what it said? Can anyone check this later? Each term here names one of those questions precisely.

Every entry says what the idea is, why it matters for AI-written changes in particular, and how Verify uses it, including the parts that are not built yet.

02How to read it

Start with running, then evidence.

Begin with base and head, control tests and mutation testing: they are about whether a change really works. Then hallucinated packages, dependency bumps and the test surface: claims a pull request makes that can be checked. Finally receipts, attestations, Sigstore and SLSA: how a result becomes something anyone can verify later.