You operate a production model endpoint with automated model-quality monitoring. After the monitoring system reports violations, you retrain the model using a dataset that reflects current production traffic, deploy the new model, and run the first monitoring job. Violations persist on the endpoint. What should you do to resolve the reported violations?

The correct approach is to recompute the monitoring baseline. When a model is retrained and deployed, its expected data distribution and performance characteristics may change. The monitoring system compares current production data against a previously established baseline. If the baseline is outdated, it will incorrectly flag violations even if the new model is performing as expected. Recomputing the baseline using the new training data provides an accurate reference for the monitor to evaluate the updated model's performance. Manually triggering the monitoring job again without updating the baseline will yield the same results. Deleting and recreating the endpoint is an unnecessary and disruptive action that won't address the underlying issue of an outdated baseline. Retraining the model again using combined data might improve model performance but doesn't resolve the monitoring system's misinterpretation of the new model's behavior against an old baseline.