The design-principle literature has four major contributors that matter for this book. They are not interchangeable; they contribute differently. Tufte's contribution is the two heuristics that ground the entire audit tradition. First: data-ink ratio , the proportion of total ink devoted to encoding data should be as high as possible. Remove ink that doesn't encode anything. Second: proportional ink , the visual area of a chart element must be proportional to the data value it represents. These are heuristics, not laws. They require judgment to apply.

Proportional ink is the most concrete of Tufte's principles, and the one with the clearest empirical grounding. Stevens' power law gives the mechanism: the eye perceives area sublinearly. When the visual area of a chart element doesn't match the data value , because the baseline is truncated, or radius rather than area is encoded, or three-dimensional perspective distorts the geometry , the reader's perception diverges from the underlying value in ways the math predicts. The violations are consistent across chart types.

Tufte's data-ink ratio , the proportion of total ink that encodes data , is a heuristic, not a formula. No one computes it literally. Its value is as a direction: when you look at a chart and there is a lot of ink, ask which of it is doing work and which is decorative. The strict Tufte reading: minimize non-data ink. The implication: remove gridlines, decorative borders, heavy axis lines, and any graphic element that doesn't directly encode a data value.

Color encoding has three distinct uses in data visualization, and using the wrong one for a given data type is a Bertin-class error. Categorical color uses distinct hues to encode categorical attributes , attributes with no inherent order. Five product lines, eight countries, four political parties. The hues should be visually distinct but similar in luminance, so no category reads as more important than the others by virtue of appearing darker or brighter.

Annotations are text or graphical elements that make specific parts of a chart legible. The correct strategy for annotations follows directly from Cairo's "the chart must answer the question" principle: every annotation should answer a question the reader might have. An annotation that answers no question is chartjunk. "What does this chart show?" The title and subtitle answer this. A title that says "Q4 Performance Highlights" does not answer the question , it names an occasion.

Stephanie Evergreen and Ann Emery's twenty-two-point checklist operationalizes the four-source synthesis into a workable audit instrument. Five categories, twenty-two items. Text (five items). Is the title clear and informative , naming the finding, not the occasion? Are axes labeled with units? Are data labels visible at the chart's intended deployment size? Do annotations answer real questions? Is body text legible at deployment size?

The most common mistake is running the audit only after a chart is built. The checklist is most valuable during design: writing the four-move prompt with the twenty-two items in mind, so that the output requires minimal correction. Consider what happens when the checklist is applied before the prompt:

You can audit any visualization using the Evergreen/Emery twenty-two-point checklist and produce a written critique with specific corrections. You know which category each item falls into, what principle it operationalizes, and what the follow-up prompt looks like for each failure. You can apply Few's clarity-over-minimization resolution to any "decoration versus function" dispute. The criterion is "does this support the message?" not "is this strictly data ink?" Functional redundancy stays. Decoration goes.

Exercise 16.1 , 22-point checklist on a published chart. (Tests: applying the full checklist) Find a chart in a recent publication , corporate report, academic paper, journalism, or government dashboard. Walk through all twenty-two items, categorized by the five checklist groups (text, arrangement, color, lines, overall). For each item, state pass or fail. Count the failures.

The design principles chapter brings the book's accumulated rules into application. The model has read every design-principles book ever published. Where the model genuinely helps: critiquing your draft chart against canonical principles , data-ink ratio, perceptual ordering, color choice , and surfacing where you violated them.

The ideas in this chapter didn't appear from nowhere. Edward Tufte published The Visual Display of Quantitative Information in 1983 , and coined the terms "chartjunk," "data-ink ratio," and "sparklines." His insistence on stripping decoration from charts continues to shape how serious data graphics get made. Edward Tufte, circa 1983. AI-generated portrait based on a public domain photograph (Wikimedia Commons).

Stephanie Evergreen and Ann Emery's twenty-two-point checklist operationalizes the four-source synthesis into a workable audit instrument. Five categories, twenty-two items. Text (five items). Is the title clear and informative , naming the finding, not the occasion? Are axes labeled with units? Are data labels visible at the chart's intended deployment size? Do annotations answer real questions? Is body text legible at deployment size?