How to approach it
Define the population, sampling window and inclusion rules. Report missing data and possible selection bias. Publish definitions, aggregate tables and the method used to calculate results.
Define what counts as an observation and what would cause you to discard it before collecting data. If you sample storefronts, record eligibility, date and region. Publishing a short data dictionary and a worked calculation helps readers interpret the findings and prevents a metric from being reused outside its intended scope.
Worked example
For a study of store navigation, specify how products were sampled, what counted as an orphan and whether private and out-of-stock listings were excluded.
How to verify the change
Keep a separate list of limitations such as incomplete HTML, unavailable products or a sample skewed towards a specific platform. Observe privacy and permission requirements for proprietary data. Include denominators with results and distinguish observed patterns from the explanation you think might account for them.
If the dataset excludes failed or unavailable pages, describe the number excluded and reason. A clean-looking result can still reflect a narrower group than readers expect. Give them enough information to judge whether it applies to their own store.