Proof of work.

AA-1, filed on ourselves first

The Agent Governance Attestation is an open standard for disclosing how AI agents are governed: which controls an organisation runs, at which tier, with what evidence. We proposed it, steward it in the open, and filed the first attestation on our own systems, which means the disclosure regime we recommend to clients is one we already live under.

We run what we recommend

The firm builds and operates a governed investment-policy runtime in production, under four rules. Language models work on the authoring path and are banned from the enforcement path, so the decision that enforces an obligation is deterministic. Every automated process runs in shadow before it is trusted, logging what it would have done while enforcing nothing. Evidence is produced by the system as it works, not reconstructed after the fact. And the customer owns every compiled artifact, exportable and independently verifiable.

The runtime behind that doctrine carries a four-figure test suite, cryptographic attestation with independent signing, trusted timestamps, and offline verification. The full operation walkthrough, including the build history and a sample examiner evidence pack, is part of the first conversation rather than a public download.

Written before it was asked for

The document repo carries the firm's working papers and a public change log of this site itself, revision by revision. A firm that asks institutions to keep living, examinable records keeps its own in the open.