Your question is Reconciling Conflicting Operational Metrics. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
You're evaluating whether a recent improvement to an autonomous driving workflow is ready to influence operational decisions, but the key data sources do not agree. Offline log analysis suggests the change reduces intervention-related events, while on-road review and simulation-based metrics show weaker or mixed results. Different stakeholders are already citing the source that best supports their position, and you need a decision process that is credible, fast, and repeatable.
How would you handle conflicting data sources when making decisions?