A customer's evaluation results show a performance gap between two demographic subgroups on an otherwise strong-performing model. What does this finding mean for the rollout?
Select an answer to reveal the explanation.
Short Explanation
Any gap you found is a gap you now own. Strong overall performance doesn't cancel it out. You need to address, assess, or mitigate the disparity before calling the model ready.
Full Explanation
The correct finding is that the disparity itself needs to be addressed, or at minimum explicitly assessed and mitigated, before the model can be considered ready to deploy at that performance level. Once a subgroup gap is identified, it is a specific, actionable finding that governance obligations expect to be assessed and addressed, not a detail that a good overall score outweighs. The option saying nothing further is needed because overall performance exceeds the customer's target is incorrect because overall performance and group-level fairness are separate properties; a strong aggregate result does not resolve a documented disparity. The option saying the smaller subgroup should be excluded from the deployment's user base to remove the discrepancy is incorrect because excluding a group from service to remove an inconvenient result is itself a discriminatory outcome, not a fix. The option saying the evaluation methodology is flawed and should be discarded in favour of an aggregate-only report is incorrect because the finding is exactly what disaggregated evaluation is designed to surface; discarding the method because it found a problem defeats its purpose. For the exam, treat a detected subgroup gap as something to remediate, not something to explain away or evaluate around.