Derive the cleaning and normalisation spec from the raw data
Inspects raw client inputs and decides the full set of cleaning rules — dedup keys, format standardisations, unit conversions, cross-source reconciliation logic — producing an explicit, reviewable spec.
In / outRaw data overview + methodology requirements → cleaning spec (dedup keys, format rules, unit conversions, integrity checks)
You might say…
“I applied cleaning rules in my head as I went and couldn't reproduce the dataset a week later — writing the spec first costs ten minutes and saves hours.”
What it does
Inspect the heterogeneous raw client inputs and decide the cleaning rules: which dedup keys, which format/naming standardisations, unit and date conversions, cross-source reconciliation rules, and integrity checks to apply. Produces an explicit, reviewable cleaning spec. Used once the inputs pass the completeness gate; the judgement layer that a deterministic transformer cannot supply for messy, unfamiliar client data.
Trigger: Use after the inputs pass the completeness gate; provide the raw data overview and the methodology's working-data requirements.
Recognise the problem?
The primitives are the commodity part. The fastest next step is a conversation about composing them into something that works for you.
Start a conversation