Priority Level
Medium
Task Summary
Replace the use of df.attrs (experimental pandas feature) for threading pipeline metadata through workflow stages. Currently original_text_column is carried via df.attrs, which is not preserved through merge/concat/groupby operations.
Technical Details & Implementation Plan
Create a PipelineContext (or similar name) dataclass wrapping a DataFrame + metadata dict. Update read_input, _run_internal, LlmReplaceWorkflow, _rename_output_columns, and _build_user_dataframe to pass/return this container instead of relying on df.attrs. Remove the .attrs comment/TODO in llm_replace_workflow.py:87.
Dependencies
No response
Priority Level
Medium
Task Summary
Replace the use of df.attrs (experimental pandas feature) for threading pipeline metadata through workflow stages. Currently original_text_column is carried via df.attrs, which is not preserved through merge/concat/groupby operations.
Technical Details & Implementation Plan
Create a
PipelineContext(or similar name) dataclass wrapping a DataFrame + metadata dict. Updateread_input,_run_internal,LlmReplaceWorkflow,_rename_output_columns, and_build_user_dataframeto pass/return this container instead of relying ondf.attrs. Remove the .attrs comment/TODO in llm_replace_workflow.py:87.Dependencies
No response