Data pipelines

Choose Clean, cast, compute or flatten, then edit its JSON configuration. Supported kinds: trim, lower, upper, replace, regex, substring, split, merge, default, cast_string, cast_number, cast_boolean, add, subtract, multiply, date, flatten. Column names are literal top-level names. Flatten uses dotted output names and rejects collisions.

{"op":"transform","transform":{"kind":"multiply","field":"price","value":1.2,"as":"price_with_tax"}}
{"op":"transform","transform":{"kind":"date","field":"date","layout":"02/01/2006","output_layout":"2006-01-02"}}

Substring offsets count Unicode characters. Arithmetic preserves decimal precision and rejects unsupported precision instead of rounding.

Drag cards to reorder them, or use the arrow buttons. Configurations use JSON: filter values keep their types, map supports constant values and copied fields, and select/drop/rename reuse response transformation rules. A pipeline supports up to 20 steps. Global operations use disk-backed SQLite; preview returns up to 20 records.

Rules use the same syntax as Data quality. An invalid output record rejects the export; no partial download is published. Final previews check only their first 20 rows. Intermediate step previews do not check final-output rules.