AUTO CDC

from pipetree.declarative.autocdc import render_sql, translate_table

flow = translate_table(config.tables["bronze.customer"], source="crm_customer_raw")
print(render_sql(flow))

For bronze.customer of the example (scd2, sequence_by: [versionnumber], delete_when: "isdelete = true", ignore_columns: [modifiedon]) this prints:

CREATE OR REFRESH STREAMING TABLE bronze.customer;

CREATE FLOW bronze_customer_flow AS AUTO CDC INTO
  bronze.customer
FROM STREAM crm_customer_raw
KEYS
  (accountid)
APPLY AS DELETE WHEN
  isdelete = true
SEQUENCE BY
  versionnumber
STORED AS SCD TYPE 2
TRACK HISTORY ON * EXCEPT (modifiedon);

It is a translation, not an executor. AUTO CDC runs only inside a Lakeflow Declarative Pipeline, where the platform owns the dependency graph. Every statement ends with ;, so rendered tables can be joined into one pipeline file. source is the upstream streaming table you name: the flow reads FROM STREAM, because Databricks requires a streaming source.

Pipetree option	AUTO CDC
business_key	KEYS
sequence_by	SEQUENCE BY
delete_when	APPLY AS DELETE WHEN
strategy: scd1 / scd2	STORED AS SCD TYPE 1 / 2
ignore_columns (scd2)	TRACK HISTORY ON * EXCEPT (...)

Only scd1 and scd2 translate. replace and append raise an error: a plain streaming table or materialized view does the job. A table with no business_key, or with surrogate_key, is refused. So is merge.snapshot: true: the flow would never delete keys missing from the batch.

Differences

	pipetree	AUTO CDC
scd2 history columns	_valid_from, _valid_to, _is_current	__START_AT, __END_AT, filled from the SEQUENCE BY values
Audit columns	_inserted_at, _execution_id, …	none
Duplicates of a key in one micro-batch	reduced to one row first	each becomes a history version
Older sequence	by sequence, when sequence_by is set	by sequence: scd1 keeps the higher, scd2 inserts an earlier version
delete_mode	soft, hard (scd1), ignore	only ignore is modeled (delete signal dropped). Other modes follow AUTO CDC’s own behavior for the SCD type: check the Databricks docs.