Document a data dictionary before importing
A data dictionary connects each column to meaning and a processing rule. Before an import, it separates unknown values, format errors and business contradictions. Start with fields that decisions depend on.
Give every field a meaning
For each column, record its name, definition, type, unit, source and contact person. A date might describe observation, publication or import: those events are not interchangeable.
Fictional example: in a contract list, “amount” is ambiguous without currency and monthly or annual frequency. Write “monthly subscription, euros, tax basis to be clarified” rather than only “number”.
Define absence without inventing a value
Write down states the source can produce: present, unknown, not applicable or unreported. An empty cell is not necessarily zero. When several codes exist, retain the original before a documented conversion.
In the example, a blank exit cost means information must be requested. Converting it to 0 would make the contract artificially cheaper. An absence may block a calculation while leaving the rest of the record useful.
Preserve identifiers and relationships
Distinguish an identifier from a quantity. Preserve leading zeros when the format requires them. Record keys linking files and treatment of unresolved references; do not join entities solely by name.
Prepare a small batch containing a missing identifier, a potential duplicate and an unresolved relation. The expected outcome can be a record awaiting review rather than automatic merging or removal.
Write checks before importing
Separate syntax, business rules and factual accuracy. A properly formatted date can be wrong; a valid address can be outdated. Write accepted and rejected examples for each check and specify the required response.
The site’s CSV checker measures empty cells, row counts and exact duplicates in one column. It does not establish identifier validity or source accuracy. Use this initial diagnosis to prepare business checks.
Version the dictionary with the file
Keep the received file, dictionary, transformations and unresolved anomalies together. Before another import, compare column meanings as well as names. An unchanged column name may conceal a new unit.
Assign ownership of meaning changes. If a file becomes incompatible, prepare a reviewable transformation on a copy and rollback evidence; do not silently replace the reference dataset.
A record to keep with the decision
| Item | Evidence |
|---|---|
| Definition | Name, meaning, unit, source and owner. |
| Missing values | Codes for unknown, not applicable and null. |
| Relations | Identifiers, keys and unresolved references. |
| Checks | Accepted/rejected examples and expected handling. |
Download the worksheet to fill in (CSV)
Frequently asked questions
Should every numeric-looking column become a number?
Not when it represents an identifier. Check meaning and preservation of leading zeros first.
Does high completeness prove quality?
It shows cells are non-empty under your chosen rule. It does not establish accuracy, currency or fitness for purpose.
Reference material
GOV.UK · Data Quality Framework
The practical checklist is an editorial synthesis to adapt to your service. It does not constitute a certification or an audit result.