Model Documentation and Governance

Every production model has a model card — a documentation standard that captures everything needed to understand, evaluate, and maintain the model. The model card contains: purpose (what business question does this model answer?), training data (what data was used, from what time period?), features and weights (what inputs drive the predictions?), validation performance (accuracy, calibration, baseline comparison), known limitations (where the model is expected to fail), monitoring configuration (what signals trigger retraining?), and version history (how has the model evolved?). The model card is reviewed at every retraining cycle. It is the institutional memory that prevents the "nobody knows how this model works" situation that arises when the original builder leaves.