Finance receives new parking citation files and must load them into an existing Delta table with repeatable, idempotent semantics. What should they use?

Finance loading new parking citation files into an existing Delta table with repeatable, idempotent semantics should use COPY INTO. COPY INTO tracks which files have already been loaded so replaying the same citation drop does not double-count fines—essential for municipal fee integrity when operators re-runnable file drops after a failed job or late file arrival. Idempotent file ingest is the operational requirement stated for the existing Delta target. Spark batch read of citation files appended to existing Delta table using streaming checkpoint misapplies the mechanism—plain batch reads have no streaming checkpoint, so reruns have no built-in loaded-file registry and still risk re-ingesting already-loaded citations and double-counting revenue and citation volumes. Delta Live Tables pipeline ingesting citation files to existing Delta table with expectations adds declarative pipeline orchestration instead of simple re-runnable file ingest, and expectations enforce data-quality rules rather than native loaded-file tracking so idempotency still depends on custom logic rather than repeatable file-level semantics. Auto Loader streaming citation files into the existing Delta table using schema inference provides continuous streaming ingest with automatic schema detection, not simple re-runnable file ingest, and schema inference handles evolving columns rather than file-level idempotency so it does not provide COPY INTO's repeatable idempotent semantics for an existing Delta target. Exam caveat: COPY INTO idempotency is file-oriented—schema evolution, corrupt files, and partial drops still need monitoring; do not confuse COPY INTO with MERGE on business keys unless you also dedupe by citation id when filenames change. Operational check: COPY INTO a citation folder into the existing Delta table, re-run the same COPY INTO, confirm row counts do not increase from duplicate files, validate a sample citation_id appears once, and only then publish finance totals.