monitoring-data-drift

What problem does it solve?

It helps you determine whether a deployed machine learning model is degrading because the input data has shifted, so you can investigate the cause before retraining.

Core Features & Use Cases

• Chooses drift metrics by feature type, using PSI for continuous and ordinal features, Jensen-Shannon for categoricals, and proportion checks for booleans.
• Calibrates alert thresholds against baseline noise, which reduces false alarms on seasonal, high-variance, or cohort-specific features.
• Produces per-feature drift tables, attribution categories, cohort breakdowns, and root-cause hypotheses for post-deployment monitoring and incident triage.

Quick Start

Ask this skill to compare a stable reference window with current inference traffic, calibrate per-feature drift thresholds, and return the top drifting features with likely root causes.