Evidence of agentic work, one action at a time.
On instrumented paths, each observed action can produce a lasting, tamper-evident receipt: the action, its declared purpose, the review result, and a commitment to exact parameters without exposing them. It gives your team evidence to review, not a legal guarantee.
A real receipt, from a fixed test fixture.
A real telos-receipt v0.1 output, minted from a fixed synthetic fixture and verified offline before publication. It shows the standalone receipt contract; it is not client work, and its own engine field says no governance engine was wired to it. The stronger evidence is an internal TELOS run, our framework observing our own agent, on The run. Download the raw JSON.
sample-receipt.json · telos/0.1-unsigned
What a hash match shows
The payload matches the recorded integrity hash. If anyone altered the read, the scores, or any payload field after the receipt was written, the recomputed hash will not match.
What it does not show
Who issued the receipt, or when. The hash is key-free, so authorship and time need a signature these standalone samples declare they do not carry. A receipt can also record a decision that stayed in scope and was still the wrong call. It tells you how the work happened. Whether it should have happened is a judgement someone has to make.
Three reads. Nothing is blocked by any of them.
These public review labels record a note for the human.
Raw schema value, shown publicly as "within purpose". The action sat close to the declared purpose. Noted, and the work went on.
Shown publicly as "question raised for review". Ambiguous against the purpose; the record carries a question for the human.
Shown publicly as "surfaced for human review". The read routes to a person, and the human decision becomes part of the record.