If you really want to do this potentially kedro hooks allow you to set things up before_pipeline_run

Try this instead (ONLY as a workaround): . # catalog.yaml: actual_data: ... filepath: .../01_raw/actual_data.csv . actual_data_updated: ... filepath: .../02_intermediate/actual_data_updated.csv . # nodes.py Node1: 1. Reads <all_data> and compares it to <actual_data> 2. calculate <new_records> 3. do the transformations on <new_records> 4. * Do some magic to avoid circular dependency * 5. return the new <actual_data_updated> . The trick on step 4 is to manually save ( pd.to_csv ) the <new_data_updated> into the same path as the <actual_data> (filepath: .../01_raw/actual_data.csv) and without stating this on the pipeline.py, so that kedro doesn't see the dependency problem. . You can do this as a workaround, but this is a very very bad practice, if you are on a hurry go ahead, and if not, is best if you plan your pipeline better