Enterprise Data Visualization

Prefer matplotlib / seaborn when

• the user is working in Python notebooks or scripts
• the goal is EDA, reporting, or static analysis output
• the chart should be easy to reproduce in analytics workflows
• the user wants something quick, readable, and dependable

Prefer plotly / altair when

• the user wants light interactivity without building custom web code
• the user is staying in Python or notebook workflows
• speed matters more than low-level visual control

If the request is ambiguous

• If it sounds like a web/HTML request, defer to D3.
• Otherwise default to matplotlib/seaborn.

Workflow

1. Identify the communication goal.
   • comparison
   • trend over time
   • distribution
   • relationship
   • composition
   • ranking
   • before/after change
2. Identify the output environment.
   • HTML/web
   • notebook
   • script
   • slide/report image
3. Identify the dataset grain and important fields.
4. Choose the chart type conservatively.
5. Explain why the chosen chart fits the task.
6. Produce implementation-ready code.
7. Apply enterprise styling and labeling defaults.
8. Include limitations or likely misreads if relevant.

Output contract

When generating a visualization, provide:

1. Chart choice
   • selected chart type
   • why it fits
   • why obvious alternatives are worse if relevant
2. Assumptions
   • inferred grain
   • missing context
   • any mock-data assumptions
3. Code
   • complete runnable code when possible
   • not fragments unless user asked for fragments
4. Styling notes
   • title/subtitle/footer guidance
   • palette choice
   • labeling/annotation decisions
5. Caveats
   • interpretation risks
   • what a human should review

Style rules

Read these references when needed:

• For overall styling: references/style-rules.md
• For choosing a chart: references/chart-selection.md
• For D3 implementation patterns: references/d3-templates.md
• For Python implementation patterns: references/python-templates.md

Required defaults

• Prefer direct labels over legends when feasible.
• Use sorted bars for ranking charts.
• Use horizontal bars for long category labels.
• Start bar charts at zero unless there is a strong reason not to.
• Use line charts for time trends.
• Use histograms or box plots for distributions.
• Avoid pie charts unless the user explicitly requests one and the number of slices is small.
• Avoid 3D charts.
• Use color sparingly and semantically.
• Highlight one key series/category and mute the rest when storytelling is needed.
• Keep gridlines subtle.
• Minimize chartjunk.

Enterprise styling posture

Target a visual feel that is:

• polished
• restrained
• executive-friendly
• presentation-safe
• similar to what people informally think of as a major-bank blue design system

Use:

• dark navy/ink for primary text
• medium blue for emphasis
• light blue sparingly
• neutral grays for scaffolding
• white or very light gray backgrounds

Do not:

• use neon palettes
• over-animate
• create infographic-like clutter
• use novelty fonts
• imitate official marketing assets

Human review triggers

Require extra human review when:

• the chart may influence a decision
• the metric definition is still disputed
• the chart includes sensitive business context
• annotations or takeaway text make claims beyond the visible evidence
• the chart may be shared externally