The gradual decay of a model's accuracy as the world shifts away from the data it was trained on.
Model drift is what happens when a model that once performed well slowly gets worse, not because the code changed but because the world did. Data drift is a shift in the inputs - new sources, changed distributions - while concept drift is a shift in the underlying relationship the model was trying to capture. Either way, a model trained on last year's world degrades against this year's.
Drift is a data problem more than a modeling one, and it is why deploying a model is the start of the work, not the end. Catching it requires monitoring live performance against fresh ground truth and retraining on current data before the decay bites. A dataset that stays fresh and well-labeled is the raw material for keeping a model honest over time.
A live evaluation measures EdgeOrigin against your coverage requirements: decision-ready real-time data, delivery latency into your systems, and record-level provenance.