The EU AI Act treats regulation as part of the infrastructure that enables safe, trustworthy and market‑ready innovation. Achieving this ambition requires that evidence generated during implementation be turned into governance and legal knowledge, which in turn supports consistent interpretation, effective oversight and adaptation as technologies evolve. The producers of such evidence and the users of it usually belong to different professional domains. To bridge this gap, we introduce MARLA (Map, Assess, Report, Learn, Adapt), a five‑stage cyclical scaffold that centres on embedding legal requirements into socio‑technical practices across local, national and European levels of the AI Act’s governance architecture. MARLA is deliberately non‑prescriptive; it offers technical and legal stakeholders a common vocabulary. The first three stages—Map, Assess, Report—each yield a documentable form of regulatory learning, which is then refined through Learn and Adapt for continuous improvement. We illustrate the scaffold with two piloted case studies and a prospective national‑to‑European scenario.
Review