Airbyte schema-change policies decide how source drift reaches the destination

The page explains per-connection settings for source schema changes. Detection runs before sync on a cadence that differs for Cloud and self-managed deployments; users can refresh manually. New fields are created in the destination; removed fields keep history until a clear/refresh; new streams backfill like an initial sync; removed streams stop updating but leave existing data.

Pipeline and governance teams need predictable behavior when upstream APIs or tables evolve. Explicit propagation rules reduce silent NULL fills, type mismatches, and failed syncs—especially where Destinations V2 records incompatible types as row errors in airbyte_meta.

Shared meaning across systems breaks when schemas drift without a contract. These controls are operational governance over the sync catalog: they decide whether destination meaning tracks the source, freezes history, or pauses for human review when primary keys or cursors change.

Cursor removal pauses the connection for manual review; primary-key changes are flagged as breaking in the extract’s truncated section, underscoring that identity fields are treated differently from ordinary columns.