| name | v1-reviewing-data-graphics |
| description | Use when reviewing charts, dashboards, or quantitative visualizations for integrity. Triggers on "review this chart", "audit this dashboard", or "misleading visualization". |
| allowed-tools | ["Bash","Read","Grep","Glob"] |
Reviewing Data Graphics
Review charts, dashboards, and quantitative displays for truth, evidence, comparison, and clarity.
This is a review-only skill: follow the shared review skeleton for posture, the findings-first output format, and severity tiers. This skill's lens is quantitative integrity (Tufte/Few) — is the display telling the truth?
Quick Start
For a chart, dashboard, screenshot, HTML report, or component:
- Identify the question the display claims to answer.
- Inspect the actual visual surface when available; do not review only source code, OCR, alt text, or data tables when the rendered graphic is available.
- Check integrity before taste: proportional representation, context, labels, units, comparison baselines, and omitted data.
- Check whether the display helps the viewer reason about the data: comparisons, detail levels, density, direct labels, and integration with surrounding text.
- Return findings first, ordered by severity, with concrete redesigns and validation checks.
What To Review
Review every visible quantitative element:
- Charts, maps, sparklines, scorecards, tables, gauges, funnels, timelines, and cohort views.
- Legends, labels, axes, gridlines, annotations, units, colors, and text around the graphic.
- Empty/loading/error states if the display is generated by an app.
- Data source notes, filtering controls, date ranges, denominators, and aggregation labels.
If the display is not visible, first try to render or open it. If rendering is blocked, say exactly what was reviewed instead.
Review Workflow
1. Establish The Graphic's Job
Extract or infer:
| Field | Questions |
|---|
| Purpose | Is the display for description, exploration, comparison, monitoring, persuasion, or decoration? |
| Audience | Who must make a decision from it? |
| Data | What are the units, denominator, grain, date range, filters, and source? |
| Claim | What conclusion does the display invite? |
| Needed comparison | Compared to what baseline, cohort, target, or prior period? |
If the display has no clear job, mark that as a finding before discussing style.
2. Integrity Pass
Read and apply the canonical visual evidence checklist, especially its integrity, comparison, data-ink, density, labeling, interaction, and validation sections.
Keep this skill focused on the data-graphics review: identify the display's job, inspect the visual surface, map checklist failures to severity, and propose redesigns tied to the viewer's decision. Do not maintain a parallel copy of the checklist here.
3. Classify Findings
For each actionable issue, classify:
- The viewer decision or claim at risk.
- The checklist category involved.
- The evidence visible on the rendered display.
- The likely misleading conclusion or interpretation cost.
- The smallest redesign that fixes the decision problem.
Avoid generic advice such as "make it cleaner" or "use better colors." Tie every redesign to the claim, data, or viewer task.
Output Format
Use the Output Format in the review skeleton; this skill adds no extra section.
Severity tiers follow the skeleton; in this skill:
- Critical: likely to cause a materially false conclusion.
- High: hides essential context, scale, denominator, or comparison.
- Medium: meaningfully slows interpretation or buries the main comparison.
- Low: polish issue with a clear readability benefit.
Residual risks to list when there are no findings: unavailable data source, uninspected responsive state, or missing source notes.
Review Checks
Before finalizing, ask:
- What conclusion would a rushed viewer take away?
- What comparison is required to believe that conclusion?
- What visual encodings are not carrying data?
- What data is missing because it would weaken or complicate the claim?
- Would a table, small multiples, direct labels, or annotations make the display more truthful?
- Did the review inspect the rendered display, not just the implementation?
Chaining
- Use
v1-html-it when the user asks to implement or package a self-contained visual artifact after the review.
- Use
v1-reviewing-usability when the main risk is interaction, task completion, state visibility, or recovery rather than quantitative representation.
- Use
v1-prd when the review reveals requirements, data contracts, states, or validation criteria that must be specified.
When Not To Use This Skill
- Do not use for general UI reviews without a quantitative display; use
v1-reviewing-usability.
- Do not use for creating a polished HTML artifact from scratch; use
v1-html-it.
- Do not use for broad product strategy or market evidence; use
v1-strategy-review or v1-learning-from-customers.
Examples
Misleading metric card
Finding: A revenue card shows +42% without baseline, date range, denominator, or absolute revenue.
Fix: Show current value, comparison period, prior value, date range, and whether the percentage is nominal, real, gross, or net.
Overdecorated bar chart
Finding: A 3D bar chart uses perspective and shadows that make similar values look materially different.
Fix: Use a flat bar chart or dot plot with direct labels and a scale that matches the decision being made.
Dense but useful report
Finding: A table-chart hybrid contains many values but supports clear scanning by row and column.
Fix: Preserve density; improve row grouping, units, and annotations instead of splitting it into many sparse cards.
