← All use cases
Knowledge
AI research
Model versions, prompts and evaluation runs, anchored before publication.
The problem
Claims about what a system produced are usually unverifiable after the fact — the model has moved on, the prompt was edited, and the output was tidied before it was shown.
01
What is anchored
Model version, prompt, evaluation run and the results, as they were produced.
02
When it is revealed
Sealed before publication, revealed alongside the paper or release.
03
Who checks it
Reviewers, replication teams and anyone citing the work.
The cycle is identical
Canonicalise the artifact, hash it with a secret nonce, anchor the fingerprint, then reveal inside the window. Only the schema and the window change from one application to the next.
See the five steps →Evaluating this for your organisation?
The whitepaper specifies the canonicalisation, the hash construction and the contract surface in full.