Record count drop-off between pipeline stages indicates data loss

Significant record count decrease between pipeline stages (e.g., source reads 10,000 records but only 500 reach sink), indicating filtering, transformation errors, or data quality issues

Monitor per-stage record counts in pipeline run details. If drop-off detected, add Error Handler node after transform stages to route failed records to separate output (BigQuery table or GCS file). Examine error records to identify problematic data values