validate-source-alignment

Data drift detection in BigQuery compares upstream source table counts and presence against ml_feature_store_v2 to identify silent data-flow failures and prevent defaults from creeping into predictions.

You can validate ETL pipeline data quality for a specific date by running deep validation, which checks history, composite factors, and Vegas data groups against the feature store for the given date.

Yes, the validation supports both quick and deep modes, running SQL-based checks against BigQuery tables to verify source alignment with the feature store across daily history and composite factors.

Operational reporting for data precision validation provides clear pass, warn, or fail statuses as ready-to-ship results, allowing you to automate alerts for upstream source and feature store misalignment.

You should run deep validation for a specific date instead of quick validation when you need end-to-end coverage of history, composite factors, and Vegas data pipelines to troubleshoot potential silent data-flow failures.

Upstream source data misaligns with the feature store in BigQuery due to silent data-flow failures, which this validation process detects by comparing source data counts and presence to prevent prediction defaults.