Overview / Resources

Resources for working with data

A glossary, a checklist for preparing an export safely and a guide to reading a change log. Free to use and share inside your organisation.

Checklist

Before you export anything

  1. Write the question the data must answer in one sentence.
  2. List the columns that question needs, and remove the rest.
  3. Remove names, contact details and free-text notes unless they are essential for matching.
  4. Note the date range and any filters applied in the source system.
  5. Record how many rows the source system says it exported.
  6. Save in CSV with a header row; avoid merged cells and colour coding as meaning.
  7. Write down any local rules people follow, such as "we leave the field blank for trade customers".
  8. Agree a secure transfer route before sending; never use an open email attachment for personal data.

Glossary

Terms in plain English

Profile
A description of a dataset's columns: types, blanks, distinct values and inconsistencies.
Canonical value
The single agreed spelling or code that variants are mapped to.
Exception list
Records that failed a rule and were set aside for a human decision.
Change log
The record of every rule applied and how many rows it affected.
Reconciliation
Checking that counts or totals match between the source and the output.
Grain
What one row represents: an order, an order line, a customer per month. Mixing grains causes double counting.
ISO 8601
The international date format YYYY-MM-DD, which sorts correctly and cannot be misread as month-first.
Data minimisation
Using only the personal data that a task genuinely needs.

Guide

Reading a change log

Each line of a change log names a rule, the column it touched and the number of rows affected. Read it top to bottom: rules are listed in the order applied, because order matters. A large count on a rule you expected to be rare is worth a question. A rule that affected zero rows can usually be retired. If totals in and out do not reconcile, the log should explain the difference line by line.

Need a practical aid? The column profiler gives you a quick first profile of a small sample.