Data dictionaries: why definitions decide the answer

A data dictionary is the document that states what each field in a dataset means: what counts as a sale, what counts as a store, when a week begins, and what happens to a return. It is the least glamorous artefact in commercial analytics and it decides more disagreements than any statistical method.

When two credible sources report different figures for the same market, the explanation is almost never that one of them is wrong. It is that they answered slightly different questions.

What belongs in one?

Every field, with a definition written in business terms rather than in database terms. What the unit of observation is — a transaction, a line item, a store-week. What is included and what is excluded. How nulls are handled. And the vintage: when the data was extracted and what has changed since.

The exclusions matter as much as the inclusions. A dataset that quietly omits a store type, a transaction class or a period is making a claim about the market by omission.

Why does the definition of a sale vary?

Because there are several defensible answers. Gross or net of returns. Including or excluding tax. Before or after a discount. Counting a voided transaction or not. Each is reasonable and each produces a different total, and a comparison across two sources that differ on any of them is not a comparison.

This is the first thing to check when a number surprises you, and it resolves a surprising share of the cases where two reports appear to contradict each other.

What about the definition of a store?

Equally variable. Does a store that opened mid-period count for the whole period? Does one that closed appear at all? Does a store that changed ownership continue as the same store, or begin again? Those choices determine whether a total reflects the market or reflects the panel.

A total that grew because stores joined is not growth, and only the definition tells you which you are looking at.

How should a buyer use a dictionary?

Read it before the first report rather than after the first dispute. Then keep it, because definitions change and a report produced under one set of rules is not comparable with one produced under another.

Ask specifically about changes: when the definitions were last revised and what moved. A provider who can answer that precisely is a provider that takes its own methodology seriously.

What does a missing dictionary tell you?

That the answers exist in somebody's head rather than in a document, which means they can change without anyone noticing. That is not evidence of bad faith; it is a maturity signal about the data operation.

For a brand making distribution or pricing decisions on a dataset, that maturity is part of what is being purchased. It is fair game to ask for it, from any provider, NRS Insights included.

How do definitions cause trouble inside a business?

When two teams answer the same question from two systems. Sales reads one definition of revenue, finance reads another, and the meeting becomes an argument about numbers rather than a decision about a market.

A shared dictionary does not resolve which definition is right. It resolves which one is being used, which is the part that was actually blocking the conversation.

What should be documented beyond the fields?

The lineage: where the data came from, what transformations were applied, and who can change them. A field definition without lineage tells you what a number means but not whether it will mean the same thing next month.

Restatement policy belongs here too. If history can be revised, the document should say under what circumstances and how far back, because an analysis built on a version of history that no longer exists is not reproducible.