Every testing framework in production answers one question: does the output match the specification? That is verification, and it is necessary. It also quietly assumes the spec is a faithful encoding of what was actually wanted.
That assumption fails constantly. A spec can pass every test it has and still be a perfect implementation of a mistake. I have shipped that mistake, with a green suite.
Telos reframes testing around a different question: does the artifact serve the intent? It treats intent itself as a first-class, versioned, testable thing, not a fixed input you assume is correct.
It models the path from desire to behavior as a chain with lossy links: real intent, then stated intent, then specification, then behavior. Most failures happen in the translation between those, not at the last step where conventional testing looks.
So there are three planes. Conformance: behavior matches spec, the only plane normal testing instruments. Validation: the spec faithfully encodes what was asked. And intent integrity: the stated intent actually matches the real intent, and is coherent, complete, and still fresh.
It runs as a fully autonomous loop that generates intent-tests, runs them, and issues verdicts. What it deliberately does not do is invent the intent it tests against. That stays human, captured at the highest fidelity I can get and put under version control so the loop has a stable referent.