Inconsistent data reliability from external feeds undermines model accuracy and creates auditability challenges. The solution requires both understanding where data comes from (provenance) and verifying its quality before it enters the model's learning process (stage gate reviews).

Why C is Correct: The ISACA AAIR data quality governance guidance identifies establishing data provenance and implementing stage gate quality reviews as the comprehensive approach to managing inconsistent external data reliability. Provenance tracking records the origin, processing history, and chain of custody of each data source, enabling quality issues to be traced to their source. Stage gate reviews enforce quality standards at defined points in the data pipeline, preventing unreliable data from advancing to model training.

Why A is Wrong: Weighting historical data over recent samples introduces temporal bias and prevents the model from reflecting current real-world conditions---the opposite of what most AI applications require. This trade-off may be appropriate in specific contexts but is not a general mitigation for inconsistent data reliability.

Why B is Wrong: Updating model versions improves model architecture and training processes but does not resolve the underlying external data quality problems. The model update cannot compensate for ingesting unreliable data.

Why D is Wrong: Reducing data source diversity sacrifices the breadth of information that diverse feeds provide, potentially reducing model performance and representativeness. The goal is to ensure consistent quality from diverse sources, not to reduce diversity.