Tufte Charts

Tufte Charts — Data-Visualization Methodology

Apply Edward Tufte's principles (The Visual Display of Quantitative Information) to design analytical charts and tables. This skill is a design methodology, not a tooling library — the rules apply regardless of whether you render in matplotlib, plotly, D3, ggplot2, Vega, or a spreadsheet. The output should reveal complexity and support comparison without clutter, and be suitable for executive review rather than presentation filler.

When to Use This Skill

Load this skill whenever you produce a chart, a small set of charts, or a dense comparison table and the user wants analytical clarity. Triggers include: "Tufte-style", "maximize data-ink", "remove chartjunk", "small multiples", "minimal/clean charts", "executive-grade", "make it analytical", or any request where the graphic must support decision-making.

The Core Discipline (apply to every graphic)

1. Maximize the data-ink ratio. Every drop of ink should encode data. Erase everything that doesn't: heavy gridlines, borders, tick marks, background fills, redundant labels, 3D effects, drop shadows, gradients.
2. Erase non-data-ink, then erase redundant data-ink. Two passes. First remove decoration; then remove repeated encodings (e.g., a value shown both as a bar length and a label and an axis tick — keep the most direct one).
3. No chartjunk. No moiré patterns, no decorative textures, no ducks (graphics that exist to look impressive), no gratuitous color. If an element doesn't change the reader's understanding, delete it.
4. Maximize data density. Show many numbers in a small, coherent space. Prefer one dense, information-rich graphic over several sparse ones.
5. Reveal complexity at several levels — broad overview down to fine structure — in a single coherent view.
6. Tell the truth. Proportional encoding (the lie factor ≈ 1). Don't truncate a bar-chart baseline; don't over-smooth or over-aggregate away the real values.

See references/tufte-principles.md for the full ruleset, the data-ink checklist, and Tufte's specific techniques (range-frame, dot-dash plot, small multiples, multifunctioning elements).

Workflow

1. Identify the comparison. Ask: what does the reader need to compare, rank, or detect? The chart type follows from the comparison, not the other way around. Use references/chart-selection.md to pick the form.
2. Choose the most direct encoding. Position along a common scale beats length beats angle/area beats color intensity. Default to this hierarchy.
3. Draft minimally. Start with data marks only — no frame, no grid, no legend. Add back only what a reader provably needs.
4. Label directly. Put labels on or beside the data (end-of-line labels, point labels) instead of forcing a round-trip to a legend. Remove the legend if direct labeling works.
5. Restrain color. Default to a single ink color plus grey for context. Add one accent color only to mark the analytically important series/outlier. Never use color decoratively.
6. Annotate strategically, not decoratively. A few words placed at the outlier or key point — explaining why it matters — outperform a caption block. Use sparingly.
7. Inspect and prune. Look at the rendered output. For every element ask "what would be lost if I deleted this?" If nothing, delete it. Check for: text overflow/wrapping, label collisions, low text-vs-background contrast, truncated baselines.

Non-negotiable defaults

• Horizontal layouts for ranked categorical data (horizontal bars) — labels read naturally left-to-right and long category names fit.
• Small multiples when comparing the same measure across many accounts, segments, time periods, or categories: a grid of identical small charts, shared scales, sorted meaningfully. This is Tufte's highest-leverage technique for comparison.
• Direct value labels on bars/points when exact values matter; omit them when the shape/pattern is the point and labels would add clutter.
• Thin lines, light context. Reference/median lines in pale grey, data lines thin and dark.
• Sort by insight, not alphabetically — rank bars by value, order small multiples by the dimension being compared.
• No legend if direct labeling is possible. No gridlines unless reading exact values off an axis is essential — and then make them faint.

Tables (Tufte treats tables as first-class)

• Use tables for exact values and dense comparison — they often beat a chart when there are few rows or precision matters.
• Right-align numbers, align decimals, use consistent significant figures.
• Minimal rules: a line under the header and one at the bottom is usually enough — no vertical rules, no full grid.
• Order rows by insight value (rank by the key metric). Put the most important column near the labels.
• For composite scores, show the raw components alongside the score so the reader can audit it.

What "done" looks like

A reader can extract the main comparison in seconds and the fine detail on closer inspection; nothing on the page is decorative; the important point is visually obvious without a caption explaining the chart; and the graphic would survive Tufte's test — "Above all else show the data."

Read references/tufte-principles.md and references/chart-selection.md for the detailed rules and per-chart-type guidance before producing the visuals.