Model calibration decays without retraining
A model-drift alarm fires on your CTR-prediction model: the daily calibration metric (predicted-vs-actual click rate) has decayed steadily over two weeks and crossed the threshold, and AUC on the live holdout dropped from 0.81 to 0.74. There was no model deploy in that window — the same checkpoint has been serving for 40 days. Dashboards: input feature distributions for 'days_since_signup' and 'top_category' shifted noticeably starting ~12 days ago, which coincides with a big marketing push that brought in a new cohort of users. Walk through how you investigate and respond.
Stop the bleeding first (mitigate), then form hypotheses from real signals. Separate root cause from symptom, communicate status as you go, and close with what prevents a repeat.