Open methodology
The methodology we productise, published openly.
AgentAudit's evaluation, telemetry and audit methodology is authored in production engineering and published as we ship. This is the library that drives the platform.
Methodology & regulatory papers
Methodology paper
32 pagesAn evaluation methodology for production agent estates
Versioned harnesses, regression discipline and severity-gated deployment — the methodology behind AgentAudit's evaluation module, distilled from production engineering at Turing.
ReadRegulatory deep-dive
24 pagesEvidence patterns for the UK AI Action Plan
Five reusable operational evidence patterns that satisfy the UK AI Action Plan's accountability expectations — for second-line, governance and risk-committee reporting.
ReadMethodology paper
18 pagesSub-period audit reporting for sectional regulator reviews
How to scope, generate and defend sub-period reports for FCA section reviews, MHRA post-market surveillance and ICO audits — without retro-fitting evidence to scope.
ReadMethodology paper
16 pagesHeld-out validation cohorts for pre-deployment certification
Why production deployment certification needs sealed cohorts, how to author them, and how to audit cohort integrity over a deployment lifecycle.
ReadTechnical blog
Technical
Why drift detection on agent estates needs behavioural metrics, not just embeddings
A practitioner's view on the limits of embedding-distance drift and the case for behavioural assertions.
ReadTechnical
Capture-time PII redaction with an auditable evidence trail
Patterns for redacting at capture rather than after the fact, and how to make the redaction itself ICO-auditable.
ReadSector
Consumer Duty good-outcomes evidencing for agent-augmented journeys
What 'good outcomes' looks like when an AI agent sits in the customer journey — and what to log to evidence it.
ReadEngineering discipline
Severity-gated deployment for agents — the 24-hour rule
Why a green evaluation suite within the last 24 hours is the right pre-promotion gate, and how to operationalise it without slowing engineering velocity.
ReadTalks & regulator engagement
Subscribe to the practitioner briefing
Quarterly methodology updates and regulator commentary, sent to your inbox.