When syncing to Snowflake via Hightouch, never use Mirror mode on a table that other processes write to: the TRUNCATE wipes everything first. Use Insert for append-only loads and Upsert for dimensional updates. Ensure the matching column is actually declared UNIQUE or PRIMARY KEY in Snowflake or record matching will silently have no options. Tune the 50k default batch down if you need per-row error visibility.

Context: Hightouch docs (Snowflake destination): Mirror mode TRUNCATEs the destination table and then COPY INTOs all model rows, wiping existing data before reload. Insert mode is faster and better for pure row additions with per-row error identification. Upsert uses MERGE INTO. The record-matching column you select must be a UNIQUE or PRIMARY KEY column in Snowflake, or it will not appear in the dropdown. Default batch size is 50,000 rows per query and is tunable.