Notes on verifiable automated decisions.

Guide · June 2026

EU AI Act Article 12 logging: an implementation guide.

Step-by-step: capture records, anchor digests, build a verifiable evidence pack for Art. 12/19 compliance.

Paper · June 2026

Tamper-evident is not tamper-proof.

Signed receipts prove who signed. A public anchor proves when — without trusting the signer.

Paper · June 2026

Agent governance prevents. It can't prove.

Guardrails stop the bad call. They can't prove, later, what your agent actually did.

Paper · June 2026

Observability shows you. It can't prove it.

Langfuse, Helicone, Arize answer "what's my agent doing?" — not "can you prove what it did?"

Paper · June 2026

The EU AI Act wants logs you can prove.

Articles 12 and 19 — and why a log you hold yourself is not evidence.

Paper · May 2026

Logs are not proofs.

The case for deterministic replay in AI agents.