Sieve / DataDESIGN EXAMPLE
Less cleanup. More clarity.
Standardise a small dataset without silently changing its meaning. Receive clean rows and a record of every change.
01Profile the data
02Apply explicit rules
03Quarantine ambiguity
04Reconcile row counts
What you provide
A CSV you may process, a data dictionary and approved rules for dates, missing values and duplicates.
What you receive
clean.csv preserving source IDs
changes.jsonl with before/after values
quarantine.csv for ambiguous rows
What counts as accepted
No silent row deletion. Every transformation is traceable to an approved rule. Ambiguous dates remain flagged. Output plus quarantine accounts for the input population.
What is outside scope
No contact harvesting, identity enrichment or inferring sensitive attributes. New business rules and fuzzy identity matching are separately scoped.
The sample is a fixture, not a benchmark result. Workflow, model version, reviewer and reproducible test report must be supplied before a verified listing can go live.