Project summary

proofbundle turns an AI-evaluation result into a portable, signed, Merkle-anchored, selectively-disclosable receipt anyone can verify offline — Ed25519 (RFC 8032), RFC 6962/9162 Merkle inclusion, optional SD-JWT (RFC 9901) selective disclosure, in-toto/DSSE export. MIT-licensed Python; the trusted core (signature / merkle / bundle) depends only on pyca/cryptography and rolls no crypto of its own. It attests authorship + integrity of a claimed result — not that the number is correct. This grant seeds an independent security audit of that core.

What are this project's goals? How will you achieve them?

Goal: one independent, published security audit of proofbundle's trusted core, so third parties can rely on it as the signature + selective-disclosure layer for verifiable AI evals. How: (1) scope a bounded review with OSTIF — verification logic, bundle parsing/canonicalization (RFC 8785), SD-JWT checks, CI/build path, threat model; (2) run the audit; (3) remediate every confirmed finding with regression + mutation tests and publish the report. The core is deliberately tiny and standards-native, which makes it cheap to review and fast to harden.

How will this funding be used?

Entirely toward the audit. At the $5,000 minimum: auditor scoping with OSTIF plus a first review pass of the trusted core and initial fixes. At the $12,000 goal: that plus broader review coverage (bundle parsing/canonicalization, SD-JWT, CI/build path) and remediation with regression + mutation tests. A parallel Foresight Secure AI Grant covers the full audit; OSTIF sources and manages the audit team. No overhead beyond the audit and maintainer remediation time.

Who is on your team? What's your track record on similar projects?

Konrad Gruszka (b7n0de), sole maintainer. proofbundle ships a deliberately small, review-first trusted core with a mutation-gated test suite, parser fuzzing, CI, a documented security policy and formal spec (SECURITY.md + SPEC.md in the repo), and PEP 740 + SLSA build provenance on every release — built to be reviewed, not just used. It already verifies real Sigstore Rekor inclusion proofs and RFC 6962 conformance vectors offline. Repo: github.com/b7n0de/proofbundle

What are the most likely causes and outcomes if this project fails?

Most likely: the audit surfaces issues that cost more than the seed to fully remediate — mitigated because the core is small, the fixes are bounded, and the Foresight track covers the larger budget. Second: low adoption regardless of the audit — mitigated by parallel work on an in-toto eval-result predicate and inspect_ai visibility that give the receipt a home. Worst case, the funds still buy a public security review of an open-source tool — a net good even with limited uptake. Honest note: the project is young (public since July 2026), so this is early-stage funding.

How much money have you raised in the last 12 months, and from where?

$0 raised externally. The project has been entirely self-funded out of pocket — covering ongoing running costs (rented GPU compute, GitHub, AI tooling such as Anthropic, and my own hardware for local development) plus maintainer time. This is the first external funding request; a Foresight Secure AI Grant application is in preparation in parallel, and OSTIF has been contacted to scope/manage the audit.