Route each lakehouse query by shape, not by sender. Signals, rules, and how to keep dashboards, batch jobs, and agents from sharing one engine.
One logical model over Iceberg and databases on three clouds. Pushdown, egress, Reflections, and where semantic federation still breaks.
Land small Iceberg micro-batches with PyIceberg and DuckDB in a serverless function. Commits, concurrency, and why Spark is the wrong default.
Move a legacy warehouse to Iceberg without downtime by virtualizing first. Consumer cutover, parity checks, and background copy without double-ETL.
