Skip to main content
OpsRoadmaps

Checklists / model-validation-and-approval

Model Validation and Approval Checklist

The gate a consequential model passes before an organisation relies on it: tiering, independent testing against written criteria, recorded approval, and an expiry date.

9 steps · 8 required · last reviewed (next review 2026-12-20) · updated 2026-09-20

Related focus areas:ModelOps

Sign in to run this checklist

Run this for any model whose being wrong has financial, safety, legal or fairness consequences — including models that involve no machine learning. Supervisory guidance on model risk management is the origin of this discipline, but the exposure it addresses is not limited to regulated firms.

The items here are about evidence and independence. A model validated by the people who built it has been reviewed, not validated.

  1. The model is on the inventory with a purpose, an owner and a consequence tier

    required

    A model outside the register is a model outside every control that follows.

  2. Validation criteria were written before testing began

    required

    Criteria chosen after seeing the results are a description of the results.

  3. Testing was performed by someone who did not build the model

    required

    Self-validation is precisely the failure this discipline exists to prevent.

  4. Known limitations and the conditions under which the model should not be used are documented

    required

    Every model has a domain of validity; leaving it unstated guarantees use outside it.

  5. The approval names the approver, the permitted use and the evidence it rested on

    required

    "Who approved this and on what basis" must have an answer before the question is asked adversarially.

  6. An expiry or revalidation date is set

    required

    Approval without an expiry is permanent by accident, which is how a five-year-old assumption keeps making decisions.

  7. Production monitoring watches the performance assumptions the approval relied on

    required

    An approval whose assumptions nobody checks is a document, not a control.

  8. A retirement path exists so a withdrawn model stops influencing decisions

    required

    Decommissioning that leaves the job running is not decommissioning.

  9. Fairness or disparate-impact testing is performed where the model affects people

    optional

    Where decisions land on individuals, aggregate accuracy can hide a distributional problem entirely.

Run it against your real gaps

A checklist is most useful when you know which controls you are actually missing. A free OpsRoadmaps assessment scores your controls with explainable findings and builds a roadmap from the results.

Assess my stack

Checklists document self-reported practice. Items were last reviewed 2026-09-20 — adapt them to your own environment.