Insights/Tag
#MLOps
Operationalizing machine learning — model deployment pipelines, drift detection, retraining, and governance at scale.
2 articles tagged
What the Approver Actually Saw Before Signoff
A credit-risk model gets promoted, the registry records the version and metrics, and six months later an auditor asks what the approver actually saw on screen before clicking approve. Nobody can answer. This is the gap between artefact state and decision state, and what to seal into a single immutable promotion event.
May 9, 2026
Freeze the Decision, Not Just the Weights
Six months after a credit-decisioning model goes live, the audit walkthrough begins. The registry has every field you'd expect: version, metrics, dataset hash, approver name. None of it answers the only question that matters: what did the approver actually see when they clicked approve? This is the gap between inventory and evidence, and how to close it.
May 8, 2026