How to Review Failed Runs in an AI-Assisted Laboratory Workflow

Distinguish technical failure, invalid input, missing source data, policy rejection, timeout and user cancellation. Do not combine unlike events into a single error count.

The definition should reflect the validated intended use and approved procedure, not whatever is easiest for a dashboard.

Keep the run identifier, timestamp, authorised user or service, input reference, workflow and model or configuration version, error message and relevant logs.

Protect sensitive information and retention obligations. Do not overwrite the failed state merely by rerunning the task.

Determine which samples, records, drafts or downstream steps may have been delayed, duplicated or altered. Separate confirmed impact from unresolved questions.

Clinical interpretation, diagnosis and treatment decisions remain outside this operational guide and require appropriately qualified professionals.

Apply the laboratory’s approved severity and escalation criteria. Assign an accountable owner and record any containment, including holding downstream work or returning to an authorised manual process.

AI-generated explanations may assist triage, but they must not approve the classification or replace technical, quality or scientific review.

Before rerunning, confirm the source data, configuration, permissions and reason for the previous failure. Link the new run to the original event.

A successful rerun does not erase the failure or prove that every affected record is correct. Reconcile outputs through the approved review process.

Look for repeated failures by input type, workflow version, integration, time or user role. Investigate patterns through change control, supplier review or corrective action as appropriate.

Qualified laboratory, quality, security and privacy personnel must retain oversight within their responsibilities.

Should a failed run be deleted after a successful rerun? No. Can AI decide whether a failure is harmless? No. Must every technical error become a quality incident? Apply the laboratory’s approved criteria and qualified review.