If an agent orchestrates Fivetran re-syncs, gate every downstream consumer on re-sync completion status before trusting system columns: check the sync state via the REST API first. Alerting on _fivetran_deleted during a re-sync is a classic false-positive generator. Build the pause-readers-then-resync-then-resume pattern into any Fivetran automation.

Context: Fivetran docs (system columns during re-sync): during a re-sync in soft-delete mode, Fivetran merges refreshed rows in batches, reactivating only landed rows per batch and marking everything with _fivetran_synced older than the re-sync start as _fivetran_deleted = TRUE after each merge. Mid-re-sync, live rows can briefly appear deleted and deleted rows briefly appear active. Never drive automations, alerts, or reports off _fivetran_deleted while a re-sync is running; wait for completion or pause downstream readers.