Build a claim-to-data register
For every published chart, record the headline claim, source dataset, extraction date, population, denominator, filters and transformations. Identify whether numbers represent observed results, estimates or modeled scenarios. Keep the raw source separate from the presentation export.
Check visual integrity
Inspect truncated axes, unequal time intervals, changing definitions and selectively omitted categories. Verify percentages against their denominators and avoid presenting correlation as causation. If two groups differ in sample size or selection, make that limitation visible.
Review accessibility and reuse
Provide a text summary of the main finding and accessible labels or a data table where practical. Preserve source attribution in social crops and slides. Do not reuse a chart in a new context without checking whether its assumptions and time frame still hold.
Set publication gates
Require a second reviewer to reproduce the key calculation and approve the wording. Keep an update log when source figures change. If a correction changes the conclusion, update the article and associated promotional material, not just the image file.
Field checklist and acceptance criteria
Create a working audit record for data visualization claims: auditing chart sources and context. Include a representative sample, the observed baseline, evidence links, the accountable reviewer and a date for the next check. Separate a configuration that appears correct from a real end-to-end test: screenshots, export IDs and reproducible steps should make it possible for a colleague to verify the result independently.
Before approving the change, verify the intended customer experience, tracking integrity, accessibility where relevant and the effect on qualified outcomes. Record the expected behaviour and the actual behaviour side by side. If the result differs, identify whether the problem comes from implementation, reporting definitions or the underlying business process rather than treating every discrepancy as a platform defect.
Worked scenario and prioritization
Imagine a small business discovers an unexplained change immediately after a campaign or website release. Start by isolating the affected audience, date range and user journey. Reproduce the issue using one known example before applying account-wide changes. Estimate exposure using the volume of affected sessions, leads or transactions, then rank the fix by customer harm and decision impact.
A practical priority order is: prevent lost or misleading customer outcomes first; repair missing or duplicated measurement second; optimize performance only after the data can be trusted. For low-volume accounts, compare multiple weeks and downstream quality rather than reacting to a single day's percentage movement. Keep a rollback path for changes that affect production.
Reporting, governance and follow-up
Report the original problem, supporting evidence, action taken, owner and follow-up result in plain language. Avoid presenting a completed checklist as proof of improved performance. If the change affects metric definitions, annotate dashboards and explain why pre-change and post-change figures may not be directly comparable.
Schedule a recheck after a meaningful observation period and after major vendor, policy or site releases. Where customer information is involved, minimize stored identifiers and restrict access to evidence. The outcome is a repeatable control that prevents the same problem from returning, not simply a one-time audit document.