Reporting playbooks

Metric Grain Before Report Reading

Define whether a metric row represents an event, entity, group, or period before interpreting totals, rates, or comparisons.

Separate weighted and unweighted averages so a report measure reflects the population and denominator the decision actually requires.

Make the denominator, eligible population, and zero case visible so rates and percentages answer a defined report question.

Distinguish no observation, an explicit zero, and an unavailable measure so charts and totals do not flatten different meanings.

Expose incomplete current and reference periods so comparisons do not turn a calendar cutoff into a false performance trend.

Keep fiscal periods, calendar labels, and comparison boundaries explicit when reports group activity by a non-calendar year.

Define cohort entry, return windows, and eligible members so retention tables and curves explain what each cell represents.

Make axis domain, scale type, tick labels, and units agree with the measure so visual distance does not invent a conclusion.

Explain ranking, tie handling, and the Other bucket so top-N charts preserve totals and do not hide meaningful categories.

Define where cumulative measures restart, carry forward, or exclude periods so running totals explain their scope.

Prove which filters affect a headline card and how its population, measure, and scope relate to nearby report visuals.

Make report drilldowns explain a bounded child population without silently changing the question, grouping, or date scope.

Make report freshness visible enough to distinguish delayed material from a current query and investigate time-sensitive discrepancies.

Diagnose report totals inflated by one-to-many joins by comparing row grain, join cardinality, and aggregation stage.

Define and verify distinct-count measures across joins, missing identifiers, filtered populations, and report groupings.

A report combines or labels values with inconsistent units, making a correct number communicate the wrong magnitude or meaning.

A pivot’s total is calculated over a different population or aggregation than readers expect from its visible cells.

A report sorts formatted date labels or grouped periods as text, making its timeline appear out of order.

A report compares a current value with a baseline drawn from a different population, definition, or context, making the apparent change misleading.

A scheduled report arrives without the filters, data timestamp, or measure context needed to interpret the snapshot correctly.