A parks department ends a failed tree-canopy experiment but auditors may still need the model history. What is the appropriate registry action?

Marking the failed tree-canopy experiment model as archived rather than deleting history needed for audits is the appropriate registry action for a parks department. Archive retires the deployable path while preserving run and model metadata auditors may need to understand why an approach was rejected, what data was used, and how public funds were spent on experimentation. Deletion would erase that institutional memory without improving safety.

Marking the failed canopy model version as archived and purging its run artifacts for audit storage fails because purging erases the metrics, parameters, and lineage auditors need to reconstruct the experiment; archiving should retire the deployable path while preserving that history, not destroy it to save storage. Deregistering the failed canopy model but retaining the experiment run notes for auditor review fails because deregistration removes the registered version and its lineage from the registry, and informal notes cannot substitute for the preserved model history auditors need to verify spending and decisions. Registering a separate duplicate model entry from the failed run to document outcome for audits fails because it creates redundant, fragmented lineage without retiring the original failed model's deployable path or marking it archived.

Exam caveat: archival policies should align with municipal retention schedules; “failed” does not always mean short retention if the experiment touched regulated data. Ensure no endpoint still pins the archived version. Operational check: set the failed experiment model’s lifecycle to Archived, confirm it cannot be selected by standard deploy jobs, retain lineage and metrics for auditors, and record the decision rationale in the parks MLOps runbook.