K-Veritas

Verifiable Research, Institutional Integrity

K-Veritas seals published results to the code that produced them, at the time it ran.

It does not say an experiment is correct. That stays with the reviewer.

It makes sure the numbers and the code cannot change afterwards, and that the record cannot be denied, cheating included.

No access to training data required.

Official implementation of Computer Science Conferences Should Require Nonrepudiable Experimental Results (Keita and Homan, NeurIPS 2026 Position Paper Track).

Workflow

  1. Instrument

    Wrap your training script with the K-Veritas observer. It captures execution metadata passively.

  2. Sign

    Experiment metadata is hashed and signed with RSA-4096 through the attestation server.

  3. Submit

    A signed PDF report is generated, ready to attach to papers or grant submissions.

  4. Verify

    Upload the PDF to a verifier. Cryptographic validation completes in seconds.

Verification as a Service

Institutions can integrate K-Veritas verification into their review pipelines using the REST API or Python SDK. Reports are validated against the public ledger automatically.

The protocol is open and non-proprietary. Any party can verify independently without contacting the original author.

Try the verifier