When adding two more data sources to the pipeline, what part is most likely to break first? Which part would you be least confident about and why?

When adding two new data sources, the integration layer—specifically the connectors and schema mapping logic—is most likely to break first. Each source may use a different format (JSON, Parquet, CSV) and have its own authentication mechanism. If the connector code is not robust, a change in the source API or schema drift can cause failures. I would also be least confident about the data ingestion step because it often involves custom code that may not handle edge cases. To mitigate, I would implement automated tests for each connector, use a schema registry to detect drift, and add retries and back‑off logic. I would also monitor the ingestion metrics to catch spikes in failures early.