How would you handle a field that changes type from integer to string or disappears entirely during the bronze-to-silver ETL phase, ensuring extraction validation without breaking the pipeline?

Schema drift is a common problem when a source changes the type of a column or removes it entirely. In a bronze‑to‑silver pipeline you can mitigate this by decoupling the raw layer from the transformed layer and by using a schema registry. First, store the raw data in the bronze layer without enforcing a schema; use a schema inference step to capture the current structure. Next, register the expected schema in a registry (e.g., AWS Glue Schema Registry) and set compatibility rules (e.g., backward‑compatible). During the silver ETL, compare the incoming schema against the registered one. If a column changes type, cast it to the target type or keep the raw value in a separate “legacy” column. If a column disappears, fill it with a default or null value. Use data‑quality checks to flag drift and trigger alerts. Tools like Delta Lake or Apache Iceberg support schema evolution out of the box, allowing you to add or drop columns without breaking downstream jobs. By keeping the bronze layer tolerant and the silver layer strict, you can validate extraction without breaking the pipeline.