What is stored in a recipe
Each step stores its operation, its options and the internal IDs of the columns it uses, plus the input header it was written against. No code is stored and nothing in a recipe file is executed: unknown operations, unknown keys and newer schema versions are rejected on import.
When you import a recipe onto a new file, columns are mapped by ID and name. Renamed or missing columns are listed, and the steps that need them stop at preflight with a clear message until you choose a column. Source files used by Append or Join are never selected automatically.
The recipe shows a revision number. Editing a step, adding, reordering, disabling or deleting all create new revisions. Undo and redo walk through them, and History lets you roll back to an earlier revision as a new one, so nothing is erased.
Reproducibility
Steps are deterministic: the same input and recipe give the same output, including seeded sampling. The run report records each step’s operation version, row counts and findings without including row values, so it is safe to share.
Questions
Where are recipes saved?
In this browser’s IndexedDB together with your files, unless Private session is on. Export the recipe JSON to keep or share it.
Can I parameterize a recipe?
Not yet. Typed parameters with defaults are on the roadmap. For now, edit the step options after importing.
What happens if a step fails?
The pipeline stops at that step, later steps show “not run”, and the last valid output stays available. Export of the final output is blocked until the recipe completes.