Governance Architecture
A compliance result is only as strong as the system that produced it. Ours is a linear, fingerprinted pipeline: every artifact is verifiable by hash, and no stage can alter what came before.
What the Architecture Serves
Measurement axis
Scores your AI system’s behavior across 8 dimensions. Buyer-facing, framework-neutral, worldwide.
Certification axis
The 9 control areas of ISO 42001 Annex A. Auditor-facing, matched to the frameworks you report against.
The bridge between them — measured findings mapped to governance controls and per-framework evidence — is the part no output tester and no ISO auditor owns. That bridge is the architecture’s reason to exist.
The assessment moves through a fixed sequence of stages, each handing off to the next by reference.
01
Scope
02
Adversarial testing
03
Measurement
04
Certification
05
The Score Card
Each stage receives only from the one before it, and forwards by reference what the next stage needs but it did not consume.
The model that attacks is never the model that judges. Nothing grades its own work.
Interpretation never rewrites the measured numbers. The separation is enforced by the pipeline, not by policy.
Every artifact carries a cryptographic fingerprint and is referenced by it, never copied around the system.
Alter any artifact and its fingerprint changes; the receiving stage detects it on verification. The trail can't be quietly edited.
Attack traces are immutable and citable by hash, and every score traces back to the trace that produced it. Evidence is a rate over N runs, never claimed as reproducible.
The Platform
The platform doesn't distinguish your own application's backend from a direct API integration. Both start work and register a webhook, and both are treated identically.
Work is enqueued and acknowledged immediately; results return through signed webhooks, with a status query as a polling fallback. A run is running or done, full or partial — a partial run is honest, not an error.
Every operation on a run checks that the caller's tenant matches the tenant stamped on the resource. One tenant's run is untouchable by another.
What a run can do is decided by an entitlement snapshot of features and limits, frozen by fingerprint at creation — so a run behaves consistently for its whole life, gated by capability, never by a tier label in a token.
Proof · Cloud
A one-time technical assessment
Scope, adversarial testing, measurement, and a Score Card for your AI system. No framework certification.
Proof · On-premise
Continuous certification
The same path, wrapped by a human attestation gate on the way in and framework matching on the way out.