tufte-viz

name	tufte-viz
description	Use when designing, critiquing, or improving data visualizations -- applies Tufte's principles (data-ink, lie factor, chartjunk, small multiples) and implements them in seaborn

Tufte Visualization Ideation

Apply Edward Tufte's principles to design clear, honest, high-density data visualizations. In this project, implement in seaborn (objects interface preferred) — never matplotlib-only, never bare pandas .plot().

When to Use

• Designing new charts, dashboards, or reports
• Critiquing or improving existing visualizations
• Reviewing notebooks or ad_hoc/ plots for graphical integrity
• Deciding between visualization approaches (e.g. small multiples vs. overlay)
• Reducing chartjunk or improving data-ink ratio

Workflow

For new visualizations

1. Clarify the data story

   • What comparisons matter? (Tufte: "compared to what?")
   • What is the single key insight to communicate?
   • Who is the audience and what decision does this inform?

2. Select approach using Tufte principles

   • High comparison need → small multiples (so.Plot().facet() or sns.FacetGrid)
   • Dense data → tables, sparklines, or stripplot
   • Time-series → line charts with minimal grid (despine, no top/right spines)
   • Part-to-whole → avoid pie charts; prefer ordered bar or table
   • Distributions → strip/swarm/box, not 3D histograms

3. Design with data-ink in mind

   • Start minimal, add only what earns its ink
   • Default to grayscale (palette="gray"); use color only to encode a variable
   • Strip default seaborn theme cruft (sns.despine(), remove gridlines, prune ticks)

4. Apply the Tufte test (see references/tufte-principles.md)

5. Implement in seaborn (see references/seaborn-patterns.md)

   • Save plots to data/ or ad_hoc/ — never leave only in notebook state
   • Pure functions: pandas for transform, seaborn for render, no side effects in logic blocks

For critiquing visualizations

1. Check graphical integrity

   • Compute lie factor if proportions look off (visual_ratio / data_ratio, target ≈ 1.0)
   • Verify baselines (bar charts must start at 0; line charts may not)
   • Look for 3D distortion, dual axes, truncated scales

2. Identify chartjunk

   • Decorative gradients, drop shadows, 3D bars
   • Heavy gridlines competing with data
   • Moiré patterns from default seaborn themes
   • Redundant legends when direct labels would work

3. Evaluate data-ink ratio

   • What can be erased without losing information?
   • What is redundant (e.g. both color AND legend AND title saying the same thing)?

4. Suggest improvements with specific before/after seaborn snippets

References

Load only what the current task needs:

• references/tufte-principles.md — core principles from The Visual Display of Quantitative Information: lie factor, data-ink, chartjunk, small multiples, integrity.
• references/analytical-design.md — extensions from Envisioning Information, Visual Explanations, Beautiful Evidence: the 6 principles of analytical design, sparklines, layering & separation, micro/macro, range-frames, causality, confections. Load when designing dashboards, dense displays, sparklines, or explanatory graphics.
• references/seaborn-patterns.md — concrete seaborn (objects interface) recipes for each Tufte principle: despine, small multiples via so.Plot().facet(), sparklines, range-frames, palette choices, theme. Load when actually writing the plotting code.

Quick Checklist

Before declaring a visualization done:

• Lie Factor ≈ 1.0 (no visual distortion of data proportions)
• Maximum data-ink ratio (every pixel earns its place)
• Zero decorative chartjunk
• Clear labeling — title, axes, units, source
• Answers "compared to what?"
• Shows causality or mechanism where relevant
• Multivariate where the data is multivariate (not over-reduced)
• Words, numbers, images integrated — not segregated
• Reveals multiple levels of detail (micro + macro)
• Layering: primary data dominates, secondary recedes
• Appropriate data density (Tufte: more numbers per unit area)
• Implemented in seaborn (objects interface preferred), saved to disk, reproducible from a script