Become a sponsor to kraxo
AI eval results need receipts
Every published AI benchmark number is an unverifiable claim — no mainstream eval
framework signs its results. proofbundle
turns an eval result into one signed, tamper-evident, offline-verifiable receipt:
Ed25519 + RFC 6962 Merkle, one JSON file, no server. On PyPI with attested releases
(PEP 740 + Sigstore transparency log).
A receipt proves who signed these exact bytes and that nothing changed since —
deliberately not that the number is true. That honest boundary is the product.
What your sponsorship funds
- Maintenance of the ~600-line auditable trusted core (mutation-gated tests, fuzzing)
- An external security review — the single most important next step for a trust tool
- The Inspect AI integration, so any eval run can emit a receipt with one hook
Try it in 60 seconds: pip install "proofbundle[eval]" && proofbundle demo
Featured work
-
b7n0de/proofbundle
Offline cryptographic receipts for AI evaluation results — Ed25519 + RFC 6962 Merkle + optional SD-JWT. Integrity, not truth
Python 2
0% towards $500 per month goal
Be the first to sponsor this goal!
$5 a month
SelectYou keep an honest verification tool alive. Thank you.
$25 a month
SelectYour name in SPONSORS.md (optional)
You fund test & mutation-gate maintenance
$250 a month
SelectLogo + link in the README sponsors section. You fund the external security review of the ~600-line trusted core.