Lakehouse Tower of Babel: Handling Identifier Resolution Rules Across Database Engines

Multi-engine lakehouse architectures using Apache Iceberg face a critical interoperability gap: SQL identifier resolution rules differ significantly across engines like Spark, Trino, Flink, Snowflake, and DuckDB. Spark preserves casing, Trino normalizes to lowercase, and Flink is strictly case-sensitive — causing tables created in one engine to be invisible or fail in another. Catalogs like Apache Polaris, AWS Glue, and Databricks Unity Catalog add another layer of inconsistency. Two composite case studies (NovaPay with Polaris, MediStream with AWS Glue) illustrate how both case-preserving and lowercase-normalizing catalogs each introduce their own failure modes. The recommended solution is enforcing a strict lowercase snake_case naming convention organization-wide, treating identifier normalization as part of the data contract, and validating cross-engine portability through CI testing.