A city must land nightly permit CSV drops for batch analytics and also capture live counter submissions from Event Hubs. How should extraction be designed?

Using batch file extraction for the nightly permit CSVs and streaming extraction for the live Event Hubs counter submissions is the correct ingestion design. Extraction mode should match source cadence: once-per-day municipal permit drops are naturally batch files, while continuous counter events on Event Hubs require streaming so the lakehouse reflects live civic counter activity without artificial delay. Using log-based CDC extraction for the nightly permit CSVs and streaming extraction for the live Event Hubs counter submissions fails because nightly CSVs are flat file drops with no database transaction log to mine, so log-based CDC adds needless complexity without improving freshness, while the streaming reader for live counters alone is correct. Using batch file extraction for the nightly permit CSVs and Capture replay extraction for the live Event Hubs counter submissions fails on the live path because Capture replay reprocesses archived Capture files with delay instead of reading the live hub directly, so you lose real-time freshness and true event semantics that streaming already carries; the nightly CSV portion is fine as batch, but the live counter path needs streaming. Using log-based CDC extraction for the nightly permit CSVs and Capture replay extraction for the live Event Hubs counter submissions misapplies both modes: log-based CDC needs a database log that nightly flat files lack, adding cost and complexity without improving freshness for the permits, while Capture replay adds archival delay to live counters that need direct streaming. Exam caveat: Batch versus streaming is driven by arrival and latency needs, not by a preference for one API. Many municipal platforms correctly run hybrid ingestion—batch for periodic extracts, streaming for event hubs—into the same medallion lakehouse. Micro-batch can approximate streaming for some hubs, but using log-based CDC for a true nightly file drop remains a design mismatch. Operational check: Land permit CSVs with a scheduled batch job after the nightly drop arrives, configure a streaming reader on the Event Hubs counter topic, verify gold or silver consumers see daily permits and near-real-time counters as expected, and avoid using Capture replay extraction for the live counters as a substitute for streaming the hub.