← Implementer
Module 4 of 11 · 15 min
Stage 3 · Turn requirements into working governance

Data governance and representativeness

Turn data-quality obligations into evidence and controls.

Learn

For relevant high-risk AI, examine data provenance, relevance, representativeness, error management, bias testing and the relationship between training, validation and test data.

The uploaded lesson pack emphasises the need to test whether data reflects the intended population and to look for proxy variables and leakage rather than claiming 'zero bias'.

Operationalise the controls with dataset documentation, owners, quality gates, change control and reproducible test evidence.

Remember
  • •Know where data came from.
  • •Representative for intended purpose beats merely 'large'.
  • •Record known limitations and mitigations.
Client practice

Review a medical dataset that is 90% drawn from one demographic and write the evidence gap and remediation action.

Evidence you should be able to produce
Dataset recordData provenanceRepresentativeness analysisBias/error test evidence
Source basis

Based on the uploaded EU AI Specialist lesson notes and checked against the consolidated EU AI Act in force on 27 July 2026. Where the source pack and current law differ, the current law wins.

← Previous moduleNext module →