TRARDI Framework
Four methodologies that govern how we arbitrate AI use cases, audit live systems, verify sovereignty, and assure systems after deployment.
All four parts are published at version 0. Each part evolves as we engage new systems and retire weak heuristics.
TRARDI Framework: a public method for AI systems assurance in four parts. Arbitration, system audit, sovereignty, post-deployment. Version 0 published 2026.
The TRARDI Framework is the public methodological backbone of TRARDI Group's assurance practice. We publish it because trust in AI systems depends on demonstrable evidence. The framework is organized in four parts. Each answers a distinct question about an AI system: should it be built, is it built correctly, can its sovereignty be verified, does it hold in real conditions after deployment.
Question it answers
Should this AI system be built?
A 5-axis weighted scoring framework to arbitrate 10 to 30 AI candidates in 5 days, with a strict kill threshold. Applied at the start of any engagement to avoid funding the wrong use case.
Question it answers
Is this system built correctly?
Six-dimension audit of an AI system in operation: context, architecture, model, controls, supervision, and evidence-governance. Produces a findings report with severity-ranked gaps and named risk owners.
Question it answers
Can the sovereignty of this system be verified?
Six-dimension assessment of effective sovereignty: hosting, dependency graph, extraterritorial exposure, access control, reversibility, traceability. Measures what holds in practice, not what is declared.
Question it answers
Does this system hold in real conditions?
Six-dimension review of a deployed system: usage reality, stability, exceptions and overrides, output quality, control integrity, governance continuity. One-off or quarterly retainer.
Every axis is rated 1 to 5, with 5 as the best outcome. Composite scores roll up to 0 to 5, converted to 0 to 100 percent for external reporting.
Findings are categorized as informational, low, medium, high, or critical. High and critical findings block production sign-off until remediated or explicitly accepted by a named risk owner.
Controls are supported by four evidence types: artefact (a document that exists), measurement (a metric that is captured), review (a human sign-off recorded), and test (an executed verification with a pass or fail).
Every finding and every remediation carries a named owner. An unnamed risk is an unmanaged risk. The framework refuses anonymous responsibility.
Trust in assurance work depends on the framework being readable by the audited organisation, not only by the auditor. We publish to make every claim we make against a system traceable to a public rubric. Clients keep the framework after the engagement ends. Competitors can adopt it. Regulators can read it. That is the point.
The TRARDI Framework is a private methodology made publicly readable. It is not a certification scheme, not an accreditation standard under ISO/IEC 17021, and not a substitute for official conformity assessment by bodies accredited under COFRAC or equivalent European authorities. It is the methodology TRARDI Group uses to deliver its engagements and to structure its opinions.
We will walk you through how the framework would apply to a specific AI system in your portfolio. No pitch.
Book a diagnostic