AAIA: Advanced in AI Audit

Training features must be produced consistently in training and inference. Reusing the same preprocessing transformers and resources prevents training-serving skew, so the model receives inputs at inference in the form it learned from during training. AWS SageMaker documentation notes that the same transformers can be used for both training and inference, avoiding duplicated preprocessing and feature-engineering logic.