Few/Tufte Design Library · 🚫 Anti-Patterns

❌ Don't

Pie Chart

Angle decoding is imprecise. Humans cannot accurately judge wedge area or arc length. Playfair's "off day" — Tufte.

Rule: angle decoding unreliable · Legend = eye travel

✅ Do

100% Stacked Bar

Length along a shared baseline. Rank clearly, read precisely. Direct labels eliminate the legend.

Tufte: bar superior to pie for part-to-whole · Few: direct label

❌ Don't

Non-Zero Baseline (Lie Factor > 1)

Truncating the Y-axis magnifies differences that don't exist in the data. The visual effect is far larger than the actual effect.

Tufte: Lie Factor = visual effect ÷ data effect · must equal 1.0

✅ Do

Zero Baseline — Lie Factor 1.0

All bars start at zero. The visual ratio between bars exactly equals the data ratio. No deception.

Framework v3: zero baseline mandatory for bar charts (not line charts)

❌ Don't

3D Bars / False Perspective

3D extrusion creates occlusion (hidden data), distorts bar lengths, and fills the graphic with non-data ink. "Computer debris." — Tufte.

Tufte: 3D = "administrative debris" · hides data · distorts length

✅ Do

Flat 2D Bars

Length on a flat axis is the most preattentively accurate encoding. No occlusion. Zero non-data ink from perspective.

Framework v3: length encoding superior to all 3D variants

❌ Don't

Color Legend

Every legend lookup forces the eye to travel away from the data, decode the mapping, and return. Multiplied by every data point read.

Few: legends impose lookup cost · eliminate when direct labelling possible

✅ Do

Direct Labels on Line Ends

Labels at the end of each series eliminate all legend lookups. Reader never leaves the data area.

Framework v3: direct label rule · dash pattern distinguishes series

❌ Don't

High-Saturation Rainbow Palette

Rainbow palettes imply an ordinal relationship between categories when none exists. High saturation creates visual vibration (the "1+1=3 Effect").

Framework v3: 1+1=3 Effect · ≤4 distinct hues · colorblind safety

✅ Do

Single Hue — Saturation Only

One hue, varying in saturation or lightness, encodes magnitude without implying categorical distinction.

Framework v3: single-hue rule · bar/main with opacity variation

❌ Don't

Heavy Gridlines (Chartjunk)

Full-width grid lines at every tick compete with the data for visual attention. They are administrative debris — structure for structure's sake.

Tufte: gridlines are chartjunk · erase everything non-data

✅ Do

Range-Frame or No Grid

The axis tick only at min and max data values. No full-width lines. The data stands alone on white space.

Tufte: range-frame · data-ink ratio maximised

❌ Don't

Dots on Every Data Point

Filled dots on every point create visual noise and imply precision in interpolated values. The line already encodes position.

Framework v3: line without dots + end-value label = sufficient

✅ Do

Line Only + End-Value Label

The polyline alone carries all positional information. One end-value label gives the final magnitude. Clean.

Tufte: sparkline principle · remove all non-data ink from line

❌ Don't

Gauge / Speedometer

Gauges waste 75% of their area on decoration. Angle encoding is imprecise. They cannot show comparison, target, or qualitative range simultaneously.

Few: gauge replaced by bullet graph · speedometer = "video game aesthetic"

✅ Do

Bullet Graph (Few, 2006)

One bar shows actual. A perpendicular line shows target. Muted bands show qualitative range. 5× more data in the same space.

Few: Bullet Graph spec — actual + target + quality range in one visual