Know what
the numbers represent.

A row might represent a person, an event, an account, or an update to an existing item. Those are different units. Counting rows without identifying the unit can produce an impressive total that answers the wrong question.

For an attendance example, one person visiting twice is two visits but one visitor. Both measures can be useful, provided the report names them correctly. Write a short definition for each measure and keep it close to the data rather than relying on memory.

An empty field is not automatically a zero. It may mean that a question was not asked, an update has not arrived, or a value is unavailable. Combining these situations can hide a collection problem or imply a result that was never observed.

Decide which missing states need to remain separate. If a report excludes incomplete records, show the exclusion and its reason. A smaller, well-described dataset is often more useful than a larger one whose gaps are concealed.

Category names, date formats, and repeated records can change how information is grouped. Record any rule used to merge categories or remove duplicates, including a way to recognize exceptions. Keep the original source separate from the prepared version where appropriate.

Preparation should leave enough context for another person to understand the result. Useful notes include the source, collection period, definitions, exclusions, and known limitations. These details also make it easier to identify what should be revised when the source changes.