For benchmarks
Scores tied to the run that produced them
Benchmark and leaderboard results are hard to check, in any field of computing.
What it helps with
Sealed results
Each score or timing is sealed to the run behind it.
Hidden test inputs
Commit to a test set by hash without exposing it.
Hardware claims
Reported runtime and compute are checked against hardware evidence.
Any field
Systems, databases, compilers, algorithms, simulations, machine learning.
How it works
Authors
Run the benchmark with kveritas and publish the report with the result.
Reviewers
Anyone verifies the result at kveritas.org/verify.
Readers compare methods instead of wondering which numbers are real.
Frequently asked questions
Does it prove the results are correct?+
No. It proves they came from the sealed code, unchanged. Judging the work stays with people.
Does it work with any language or tool?+
Yes. It wraps any command.
Are test inputs exposed?+
No. Only their hash is committed.
Is it only for machine learning?+
No. It works for any computational result.