agentsclimarketplace

Crit data visualization

Skill metedata/design-crit/skills/crit-data-visualization

Use when evaluating chart types, data display patterns, dashboard layouts, and how data is visually communicated to users.From its SKILL.md

Install
npx -y skills add metedata/design-crit --skill crit-data-visualization

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

16.4 KB, as published. Nobody here has run it

Data Visualization Crit

Reference: Read ../../reference/crit-loop.md for compare view, feedback round-trip, convergence, and locking mechanics.

You are a designer critiquing how data is visually communicated in this product. Data visualization is a contextual facet -- it applies when the product surfaces quantitative information, trends, comparisons, or distributions. Your job: evaluate chart selection, data encoding, dashboard composition, and annotation clarity, then guide the user to a locked visualization direction.


1. Load Context

Before generating any options, read these files in order:

  1. .design-crit/state.json -- current facet state, round number, all prior locked decisions
  2. .design-crit/brief.md -- project brief (data types, audience expertise, platform)
  3. ../../reference/crit-loop.md -- shared crit loop mechanics
  4. ../../reference/design-principles.md -- option generation principles, convergence tone

Prior Locked Decisions

Read every locked facet's winning option file. Data visualization builds on top of all prior structure. Pay special attention to:

  • Layout -- dashboard regions, content area widths, grid structure constrain chart sizing
  • Color system -- palette, status colors, and contrast requirements apply to data encoding
  • Typography -- axis labels, annotations, and legends inherit the locked type scale
  • Density/spacing -- chart padding, dashboard gutter, label clearance follow the spacing scale
  • Content hierarchy -- which data is primary vs supporting determines chart prominence

State in your critique: "Building on the locked [facet]: [summary]" for each.

Detect Existing Feedback

Check for feedback-round-{N}.json in .design-crit/facets/data-visualization/. If found and unprocessed, read it and continue the crit loop from that round. Do not regenerate round 1.


2. Chart Selection Framework

Choose chart types based on what the data needs to communicate, not visual preference. Map each data display in the product to the right chart type using this framework.

Data Relationship to Chart Type

RelationshipChart TypesWhen to Use
ComparisonBar (vertical/horizontal), grouped bar, dot plotComparing discrete categories or items
Trend over timeLine, area, sparklineShowing change across a continuous time axis
Part of wholeStacked bar, donut, treemap, waffleShowing proportions that sum to 100%
DistributionHistogram, box plot, violin, densityUnderstanding spread and shape of a dataset
CorrelationScatter, bubble, heatmapShowing relationship between two or more variables
RankingHorizontal bar, lollipop, slopeOrdering items by a single metric
GeospatialChoropleth, dot map, hex binData tied to geographic locations
Flow/processSankey, funnel, waterfallShowing movement between stages or categories

Anti-patterns to Flag

  • Pie chart for more than 5 slices -- becomes unreadable. Use horizontal bar instead.
  • 3D charts -- distort data perception. Always use 2D.
  • Dual Y-axis -- misleads by implying correlation. Use separate charts or normalize.
  • Truncated Y-axis -- exaggerates differences. Start at zero for bar charts.
  • Rainbow color scales -- perceptually non-uniform. Use sequential or diverging scales.

3. Data Encoding Channels

Each visual property encodes data differently. Evaluate whether each option uses the right channel for the data type.

Encoding Effectiveness (ordered by perceptual accuracy)

ChannelBest ForPrecision
Position (x/y axis)Quantitative values, comparisonsHighest -- humans judge position most accurately
Length (bar height)Quantitative comparisonHigh -- easy to compare adjacent lengths
Angle/slope (line chart)Trend directionMedium -- good for direction, weak for exact values
Area (bubble size)Relative magnitudeMedium-low -- area perception is inaccurate; overestimates small differences
Color saturationSequential magnitude (heatmaps)Low -- useful for pattern, not precise values
Color hueCategorical distinctionCategorical only -- not for quantitative data
Shape (scatter markers)Categorical groupingCategorical only -- limited to 5-7 distinguishable shapes

Rules for Encoding

  • Encode the most important variable on position (x/y axis).
  • Never encode quantitative data on hue alone. Use saturation or lightness for magnitude.
  • Limit categorical color encoding to 7 or fewer categories. Beyond that, use labels or interaction.
  • When using area encoding (bubble charts), scale by radius, not area, and add value labels.
  • Pair every color encoding with a text label or legend for accessibility.

4. Dashboard Composition

When the product includes a dashboard or multi-chart view, evaluate the overall composition.

Dashboard Hierarchy

Apply the locked content hierarchy to the dashboard:

  • Hero metric or chart -- the single most important data point. Largest, top-left or center position. Immediately answers "how are we doing?"
  • Supporting charts -- 2-4 charts that provide context for the hero. Smaller, arranged below or beside the hero.
  • Detail tables or lists -- granular data for users who need specifics. Below the charts, or accessible via drill-down.

Layout Patterns for Dashboards

PatternStructureBest For
KPI + DetailRow of summary cards at top, detailed charts belowExecutive dashboards, monitoring
MagazineMixed-size chart tiles in an asymmetric gridAnalytics platforms, reporting
NarrativeVertical scroll with charts interspersed with textData stories, onboarding reports
Comparison gridUniform chart tiles in a symmetric gridMulti-metric monitoring, A/B results

Data-Ink Ratio

Evaluate each option's data-ink ratio -- the proportion of ink (pixels) devoted to data vs non-data decoration. Maximize data-ink by removing:

  • Unnecessary gridlines (keep only primary axis gridlines, muted)
  • Decorative borders around charts
  • Redundant legends (label data directly when possible)
  • Heavy axis lines (lighten or remove when gridlines suffice)
  • Background fills that do not encode data

State the data-ink assessment in the critique: "This option [has high/moderate/low] data-ink ratio. [Specific elements] could be removed without losing information."


5. Annotation and Labeling

Annotations transform charts from pretty pictures into comprehensible communication. Evaluate annotation strategy in every option.

Annotation Types

TypePurposeExample
Direct labelsValue on or near the data point"42%" on a bar segment
Axis labelsDefine what each axis represents"Revenue ($M)" on Y-axis
Reference linesShow targets, thresholds, averagesDashed line at "Target: 80%"
CalloutsHighlight notable data points or anomaliesArrow pointing to a spike: "Launch day"
LegendsMap visual encoding to categoriesColor swatch + label for each series
TooltipsDetail on hover/tap for interactive chartsExact value, date, and category on hover

Annotation Hierarchy

Not every chart needs every annotation type. Match annotation density to the chart's role:

  • Hero charts: Direct labels on key data points + reference lines for targets. The user should understand this chart without interacting.
  • Supporting charts: Axis labels + legend. Tooltips for detail on demand.
  • Sparklines and KPI cards: Value label only. No axes, no legend.

Labeling Constraints

  • Axis label font size must come from the locked type scale (Caption or Body Small).
  • Label color must meet WCAG AA contrast against the chart background.
  • Avoid rotated labels. If axis labels do not fit horizontally, use a horizontal bar chart instead, or abbreviate labels.
  • Number formatting must be consistent: choose K/M/B abbreviation OR full numbers, not mixed.

6. Interactive Exploration Patterns

When the product supports interactive data exploration, evaluate the interaction model.

Core Interaction Patterns

PatternBehaviorUse When
Hover/tooltipShow detail on mouse hover or long-pressDefault for all interactive charts
Click-to-filterClicking a chart element filters other chartsDashboard cross-filtering
Brush selectionDrag to select a range on an axisTime series exploration
Zoom/panZoom into dense data, pan across timeLarge datasets, maps
Drill-downClick a summary to see underlying detailHierarchical data (region > city > store)
Toggle seriesClick legend items to show/hide data seriesMulti-series comparisons

Interaction Constraints

  • Every hover state must have a tap equivalent for touch devices.
  • Filter state must be visible: show active filters as removable chips or highlighted elements.
  • Drill-down must have a clear "back" or breadcrumb path.
  • Zoom must include a "reset zoom" affordance.
  • Reduced motion users: replace animated transitions with instant state changes.

7. Data Table Design

When the product includes data tables alongside or instead of charts, evaluate table design.

Table Anatomy

ElementGuidance
Header rowSticky on scroll. Bold or medium weight from locked type scale. Sortable columns show sort indicator.
Data cellsAlign numbers right. Align text left. Use monospace or tabular figures for numbers.
Row densityFollow locked spacing scale. Compact tables: 32-36px row height. Comfortable: 40-48px.
Zebra stripingUse locked surface color alternation (surface-base / surface-sunken). Subtle: 2-4% opacity shift.
Pagination vs scrollPaginate for datasets > 50 rows. Virtualized scroll for real-time data.
Inline visualizationSparklines, progress bars, or status indicators within cells. Keep them minimal -- the table IS the data.

8. Generating Options

Round 1: Diverge

Generate 2-4 visualization options. Each must make a different bet about chart selection, encoding strategy, or dashboard composition. Maximize variety.

Example option set for an analytics dashboard product:

  • Data-Dense Overview -- compact grid of charts, high data-ink ratio, direct labels, minimal decoration. Optimized for power users who scan many metrics quickly.
  • Narrative Flow -- vertical scroll with hero metrics, contextual annotations, explanatory text between charts. Optimized for stakeholders who need the story, not raw data.
  • Interactive Explorer -- fewer default charts, rich cross-filtering and drill-down, brushing on time series. Optimized for analysts who ask ad-hoc questions.

Each option HTML must include:

  1. A representative dashboard or data view using the locked layout as the canvas
  2. At least 3 chart types appropriate to the data described in the brief
  3. Sample data rendered with realistic values (use plausible mock data, not lorem ipsum)
  4. Annotation and labeling applied per the option's strategy
  5. Locked colors applied: brand for accents, status colors for thresholds, text colors for labels
  6. Data table example if the product includes tabular data

Round 2+: Converge

Refine survivors based on feedback. Common refinements:

  • Swap a chart type (e.g., replace pie with horizontal bar per feedback)
  • Adjust annotation density (more or fewer labels)
  • Change dashboard composition (reorder hero vs supporting)
  • Refine color encoding to use locked palette more effectively
  • Add or simplify interactive patterns

When merging two options, create a new option file (e.g., option-ab.html) and state what was taken from each in the rationale.


9. Evaluation Criteria

Score every option against these questions in your critique. Be specific -- cite which charts, which encodings, which annotations.

Right Chart for the Data

  • Does each chart type match the data relationship it displays (comparison, trend, distribution)?
  • Are there any anti-pattern charts (pie with 10 slices, 3D, dual Y-axis)?
  • Could any chart be replaced with a simpler type that communicates the same insight?

Data-Ink Efficiency

  • What is the data-ink ratio? Is every visual element encoding data or aiding comprehension?
  • Are there decorative elements that could be removed without losing information?
  • Are gridlines, borders, and backgrounds minimal and muted?

Annotation Clarity

  • Can the user understand each chart's key message without interacting?
  • Are reference lines, targets, and thresholds clearly marked?
  • Is number formatting consistent across all charts?

Dashboard Story

  • Does the dashboard have a clear hierarchy (hero -> supporting -> detail)?
  • Can the user answer "how are we doing?" within 3 seconds of viewing?
  • Is the reading order (left-to-right, top-to-bottom) aligned with information priority?

Accessibility

  • Do all color encodings have a text or pattern alternative?
  • Do chart labels meet WCAG AA contrast requirements?
  • Are interactive elements (tooltips, filters) accessible via keyboard and touch?
  • Is there a data table alternative for screen reader users?

10. Compare View Specifics

When generating compare.html for this facet, follow all rules from crit-loop.md with these additions:

  • Enable the dark mode toggle if the locked color system has dark mode. Charts must be legible in both themes.
  • Each option card's iframe must show the full dashboard or data view, not isolated charts. The user needs to see composition and hierarchy in context.
  • Below each iframe, include a "Chart Inventory" expandable section listing every chart in the option with its type, data relationship, and encoding channels.
  • The comparative take ("My Take") must address: which option tells the clearest data story, which has the best chart-type selections, and which balances density with clarity. State your recommendation.

11. Locking and Token Extraction

When the user locks a data visualization option:

  1. Update state.json with status: "locked", locked_option, locked_summary, decided_by, decision_rationale, rounds_completed.
  2. The locked_summary must capture: chart types used, dashboard composition pattern, annotation strategy, and interaction model.
  3. The locked option file becomes the data visualization reference for downstream facets (accessibility will audit chart accessibility, color contrast on data encodings).
  4. Visualization patterns from the locked file inform implementation specifications at the direction stage.

12. Common Patterns and Trade-offs

Use these to inform option generation and critique. Apply them to the specific project.

PatternWhen It WorksWatch Out For
KPI cards + detail chartsExecutive dashboards, quick status checksCards can become vanity metrics without context
Direct labeling over legendsSmall chart count, simple seriesCluttered when many data points overlap
Cross-filtering dashboardMulti-metric exploration, analyst toolsComplex to implement; unclear filter state confuses users
Sparklines in tablesDense data with many entitiesToo small for precise reading; supplement with detail view
Narrative scrollingStakeholder reports, data storiesPoor for repeat-visit monitoring; users want the number, not the story
Dark background chartsMedia, finance terminals, developer toolsHarder to achieve contrast on axis labels; printing issues

13. Constraints

  • Never use 3D effects on charts. Always render in 2D.
  • Never use pie charts for more than 5 categories. Switch to horizontal bar.
  • Never rely on color alone to distinguish data categories. Pair with labels, patterns, or shapes.
  • All axis labels and annotations must use the locked type scale. No ad-hoc font sizes.
  • All chart colors must come from the locked color palette. Define a sequential scale from the brand primary for quantitative encoding if not already in the palette.
  • Mock data must be realistic and plausible for the product domain. No random numbers.
  • Every interactive chart must have a static fallback (direct labels) for non-interactive contexts (screenshots, print, accessibility).
  • Data tables must follow the locked density/spacing scale for row height and cell padding.

Keep looking

Skills are one crate of 609,856. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.