The Science of
Certainty.
K-Veritas is an open protocol that produces cryptographically signed reports certifying the authenticity of computational experiment results without accessing training data.
K-Veritas is an open product that produces cryptographically signed reports certifying the authenticity of machine learning experiment results without accessing training data.
Research output is scaling faster than trust infrastructure. Reviewers, funders, and partners need independent evidence that published results match what was actually executed.
What We Build
Experiment Verification
Wrap an ML experiment to produce a cryptographically signed, tamper-evident PDF that binds a published result to the exact code, hardware, and time that produced it. Includes source-code integrity, a source bundle, and the HMCA consistency analyzer.
Compute-Cost Attestation
Prove the declared training work was physically performed on the reported hardware. K-Veritas checks the declared FLOPs against three physical bounds, time, energy, and memory, so a hard violation is a non-deniable physical impossibility.
Agent Non-Repudiation
Record a tamper-evident, hash-chained log of an AI agent's actions. A chokepoint records every designated action before it takes effect, so no party can later deny or alter what an agent did, with guarantees of existence, content, and order.
How It Works
From experiment to verified submission in four steps.
Run Experiments
Prefix your training commands with kveritas run. Your code runs exactly as before.
Seal the Report
Metrics, hardware data, and source hashes are collected and signed with a key only the server holds.
Submit with Paper
Attach the signed PDF alongside your paper and optionally the source code bundle.
Reviewer Verifies
Upload to the web verifier for instant crypto status, metrics, and consistency checks.
Our Approach
Observers capture execution metadata from computational experiments without accessing proprietary input data. Each report is sealed through cryptographic signing and remains verifiable by any external party.
Standardization
Strengthen open protocol requirements and public verification tooling.
Adoption
Expand pilots across varied computational domains and software ecosystems.
Governance
Independent oversight with open community participation and no investment-based control.
What is written in the paper should be what was truly gotten from the experiments.
Founder, Mamadou K.
Secure the integrity of your research today.
K-Veritas is an open-source, non-profit standard maintained by the research community.
Get Involved