Research

Whitepapers, benchmarks, experiments.

Our published work on canonicalization, transparency logs, and AI assurance primitives, grounded in the problems regulators will ask us to solve before they know to ask.

If a claim can't be verified without our cooperation, we treat it as unfinished work.

What we publish

We prioritize releasing authentic work rather than aspirational claims. We prioritize reproducible test vectors and reference implementations over marketing artifacts. Our research is the foundation for the claims Veridra makes in due diligence and regulatory conversations — if the research doesn't hold up, the product can't hold up.

Focus areas

Cryptographic determinism testing

RFC 8785 canonicalization conformance vectors across SDK languages. Cross-language payload signature verification (Python/Go parity). Published golden vectors, fuzzing harness, and nondeterminism bug bounty scope.

Transparency log witness economics

Trillian-backed log trustworthiness analysis. Incentive structures and co-signing topology research. Sigstore-compatible witness cost modeling for enterprise scale.

Evidence pack compression

Structure-aware compression for signed decision streams. Inclusion-proof deduplication techniques. Verifiable summary proofs for large-scale auditing.

Drift-as-evidence statistical foundations

Behavior-change detection as compliance artifact. Published test statistics and false-positive economics. Signed documentation of drift findings suitable for regulator timelines.

Agent decision chain replayability

Canonical representation of multi-step tool calls and retrievals. Regulator-executable replay capability with cryptographic confidence that the replay reflects the original execution.

Community collaboration

Reference implementations are open source with conformance vectors versioned in public GitHub repositories. External implementations integrate into the same conformance suite as internal SDKs — if your team writes a Rust or Elixir canonicalizer, the same test vectors validate it.

Research → platform
Primitives inform capabilities
Canonicalization conformance feeds the SDK test harness. Witness economics drives co-signing policy in Attest. Drift statistics drives signed-incident thresholds in Watch. The research is not marketing — it's where product capabilities come from.
The verification discipline
Published, falsifiable, replayable
Every research claim we publish is designed to be reproduced by anyone with the same inputs and tools. That's the discipline that separates cryptographic evidence from cryptographic theater: a third party should be able to run our math and either confirm or falsify our claim, without needing our cooperation or our trust.