Published Updated

nomue verifier Release 1 candidate is now public

An experimental local verifier is available for anyone to inspect and test; it is now pinned to nomue Protocol Release 1.

What is available

September 1 update: the release candidate is now also available as the public @licklider/nomue-verifierpackage. See the npm availability notefor the direct install path, tested environments, and current release boundary.

September 18 update: verifier 0.2.1-rc.1 adds a bounded Student-t precision repair without changing the Release 1 Protocol identity or supported scope. See the implementation note.

The nomue verifier repositorycontains an experimental verifier that anyone can run on their own computer to check the supported parts of a nomue Record—a machine-readable file that connects research outputs to their evidence. It works independently of the organization that created the Record.

It checks whether the Record has the required structure, whether its content has changed, whether the set of rules it claims to follow is supported, and whether its stored Welch test results match a fresh calculation for the Release 1 support target. It does not contact a Licklider server or run code embedded in the Record.

Release status

nomue Protocol Release 1 was published on August 24, 2026.

The verifier repository is now pinned to the published Release 1 identities. The verifier package itself remains experimental at 0.2.1-rc.1; publication of the Protocol does not turn this package into a separate stable verifier release.

See nomue Protocol Release 1 Public Draftfor the current Protocol release, its scope, and the signed release artifacts.

Try the two public fixtures

git clone https://github.com/licklider-ai/nomue-verifier.git
cd nomue-verifier
npm install

npm exec -- nomue verify records/valid.json --format json
npm exec -- nomue verify records/invalid-result-mismatch.json --format json

The first fixture is expected to pass every check supported by this verifier version. The second deliberately contains a result mismatch and is expected to fail the relevant check.

What the verifier does not prove

Even when every supported check passes, the verifier does not prove that:

  • the study or its scientific conclusions are valid;
  • the input data and researcher statements are true;
  • cause-and-effect claims are justified;
  • results are correct under versions or settings outside the supported Release 1 bundle; or
  • the Record came from Licklider’s production system.

The verifier does not return a single overall VERIFIED label. Read its JSON report to see the result of each check and the limits of what was tested.