Snowflake and SQE read each other's tables
A live run against Snowflake's Horizon catalog: each engine writes Iceberg tables the other reads, both run DML on the other's tables, and 40 reads are compared value by value. All 40 match in every SQE mode. Getting there took eight fixes, among them a distributed read that put one column's values under another column's name and a parquet decoder that could not apply Snowflake's deletion vectors.
Read