Skip to main content

COMPEL Glossary / continuous-delivery-ml

Continuous delivery (ML)

Automated, governed promotion of models through lifecycle stages — development, staging, production — with gated checkpoints (evaluation thresholds, bias checks, cost thresholds, human approval where required).

What this means in practice

Aligns with ISO/IEC 42001:2023 Clause 8.1 operational-planning obligations; each gate produces audit evidence.

Synonyms

CD for ML , ML CD , governed model promotion

See also

  • Continuous integration (ML) — Automated test and build of model code, data contracts, and training scripts on every change — extended from software CI with data-schema validation, model-schema validation, and lightweight training smoke tests.
  • Pipeline — An automated execution graph connecting data ingestion, feature engineering, training, evaluation, and deployment stages — parameterized, versioned, and re-runnable.
  • Technical documentation (Annex IV) — The mandatory documentation that the provider of a high-risk AI system must draw up before placing it on the market.