agentleFS
jpoindexterjpoindexter/tufte-skills/skills/tufte-chartjunk/SKILL.md

tufte-chartjunk

Skill1 starsChanged 4 months agoMITView on GitHub

Use when a chart, dashboard, or data graphic looks busy, shimmery, or over-decorated — crosshatch or pattern fills, heavy or doubled grid lines, 3-D extrusion, fake perspective, pictorial fills, drop shadows, or unexamined software-default styling — or when auditing any graphic before publication for marks that carry no data.

The file

We split this skill into the rules it gives an agent

12 rules. About 0 are common, 12 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈12

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Displaced trust: When the rendering shows off, attention shifts from message to medium — but a graphic's credibility rests on the data, not …

    line 138
  2. Distinctive:

    Every chartjunk element lowers the data-ink ratio, so a chartjunk audit *is* a data-ink audit:

    line 141
  3. Distinctive:

    Run before publishing any chart.

    line 160
  4. Distinctive:

    If the tool emits crosshatch by default, override it in the rendering layer — never accept default fills.

    line 182
  5. Distinctive:

    Delete the grid. If the chart still reads, leave it deleted.

    line 185
  6. Distinctive:

    If look-up is required, set the grid to ~10–20% gray at ~0.25pt against ~1pt data strokes (≈ 20–30% of the data weight).

    line 186
  7. Distinctive:

    Never plot on graph paper printed-side-up.

    line 188
  8. Distinctive:

    Remove everything that does not encode data.

    line 191
  9. Distinctive:

    If the structure itself is the decoration (the fill is the message), redraw from scratch as a flat chart — or a table.

    line 193
  10. Distinctive:

    When the data is fewer than ~10 numbers and the story is comparative, ship a table, not a chart.

    line 194
  11. Distinctive:

    No information, discovery, wonder, or substance is ever generated by chartjunk.

    line 200
  12. Distinctive:

    Do not solve a visual problem by cutting content-resolution — that moves the display the wrong way.

    line 210

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Chartjunk

Overview

Chartjunk is any graphical element that does not communicate data — it burns ink, area, and viewer attention while adding nothing to meaning. Tufte sorts it into three kinds: vibrations (optical interference from busy patterns), grids (reference lines that overpower the data they support), and ducks (decoration that becomes, or buries, the data). The governing law is that a graphic succeeds or fails on its content, gracefully displayed: ornament can amplify a dull graphic's failure but has never rescued a thin data set from its own emptiness. The practical test is therefore subtractive — strip every mark that does not encode data or serve as a minimal navigational aid, then see what is lost.


§1. Vibrations — Optical Interference from Busy Patterns

A vibration is a perceptual artifact: dense, regular, high-contrast patterns applied to a graphic make the eye see shimmer, motion, or depth that is not in the data. The dominant source is moiré — two or more fine regular patterns (crosshatching, diagonal hatching, dot screens) overlaid or printed close together produce an unstable, eye-straining flicker. Moiré is unintentional optical art: the same effect 1960s Op artists pursued on purpose, smuggled into statistical graphics by default fill patterns.

Mechanism

The visual system reads closely spaced alternating lines as movement or relief — a low-level artifact, not information. Adjacent active lines also imply forms that were never drawn (the perceptual "1 + 1 = 3 or more" effect Tufte names in Envisioning Information): the more active marks crowd together, the more spurious structure the eye invents. The result is a raised visual noise floor that contaminates the whole graphic and makes the underlying quantities harder to read.

Tufte's moiré surveys (pp. 109–112)

Two samples, one finding. Random samples (1980–82) from the ten most-cited scientific journals put moiré vibration at 2% of graphics in Biochemistry, rising to 17% in Science and 21% in the New England Journal of Medicine (p. 110). A parallel survey of graphics manuals and software documentation runs worse: from Brinton's 12% up through Spear's 46%, Tell-A-Graf's 53%, and SAS/GRAPH's 68% (p. 112). The takeaway is structural, not anecdotal: when the manual or the tool ships vibration as the default, vibration is produced at scale by people who never chose it — the software-documentation peak marks exactly that. Even a statistics journal institutionalized it: JASA's own style sheet displayed a figure "prepared in the proper form" that needed 131 line-strokes plus 15 digits to convey trivial information — vibrating, unevenly spaced, misaligned (p. 110). Tufte also notes vibration may well peak for equally spaced bars (p. 109) — precisely where default fills land.

The Bertin rebuttal

Jacques Bertin argued a designer could court controlled moiré — flirt with ambiguity without surrendering to it. Tufte rejects this for data graphics: moiré is an undisciplined ambiguity with an eye-straining, illusive quality that pollutes the whole image, and there are no good examples of statistical graphics that gain anything from it. For data, vibration is always cost without benefit.

Do / Don't: Vibrations

DoDon't
Solid fills, distinct grays, or open white to separate areasCrosshatch, diagonal hatching, or dot screens on any fill
Differentiate series with a lightness ramp (white → light → mid → dark gray → black)Dense alternating line patterns at any scale or zoom
Choose hue/saturation steps (print-safe) for categoriesOverlapping screens that produce moiré at print resolution
Proof fills at the actual output resolution before committingAccept software default fills — most generate vibration

Failure mode — "we used a computer to build a duck": the rendering tool fills bars with ordered crosshatch by default, producing optical vibration and signaling technology over content. The fix is to turn the pattern off, not to pick a different pattern.


§2. Grids — Reference Lines That Overpower the Data

The grid is the most sedate chartjunk, but a heavy grid is still chartjunk. A grid has exactly one legitimate job: helping a reader look up or interpolate values. When grid lines are darker, heavier, or denser than the data, figure and ground reverse — the grid becomes the graphic and the data recedes. Tufte's rule of thumb: grids are for the working stage, plotting data at home or in the office; in finished print they should be muted or suppressed.

Grid weight hierarchy (best → worst)

Grid treatmentEffect on figure/groundVerdict
No grid; data-derived tick marks onlyData dominantPreferred for most charts
Light gray grid, weight well under the data strokesData dominantAcceptable when look-up is the task
White grid (gaps inside filled bars / erased to white)Data dominantElegant for bar charts and histograms
Mid-gray grid, equal weight to the dataCompetingMarginal; suppress if you can
Dark/black grid at full weightGrid dominantChartjunk — suppress
Doubled grid lines (box frame + inner grid)Grid dominantSevere: in the Tukey multiwindow plot, doubled lines consume 18% of the display's area and spawn optical white dots at intersections (p. 114)

Named example: Marey's train schedule

E. J. Marey's Paris–Lyon timetable (a Tufte touchstone) is extremely data-dense: every diagonal is a train, every crossing a stop. The grid must not compete with that thicket of lines. Three treatments, ranked:

Grid treatmentResult
Heavy black grid (as often reproduced)Grid dominates; trains hard to trace
Thinned black gridSlightly better; trains more legible
Gray gridBest; trains read as foreground, grid recedes to a reference layer

The correct fix is always a gray grid, never a thinned-black one. Most ready-made graph paper is printed in a dark grid; plot on the reverse (unprinted) side so the lines show through faintly without cluttering the data — and if the paper is heavily gridded on both sides, throw it out (p. 116).

The one-question grid test (SWFE p. 134)

Seeing With Fresh Eyes compresses the whole grid argument into a question to ask of any display: what is the strongest visual element here? Tufte's gloss — the correct answer is never "the grid lines." If the honest answer is the grid, the frame, or the furniture, the figure/ground relation has already inverted and no amount of restyling will fix it; the grid must recede or go. He demonstrates the repair by re-setting a plotchart with ghost grids — a grid present enough to support look-up, faint enough that the data is unambiguously the loudest thing on the page.

The look-up exception

When a graphic functions as a look-up table — readers will read off specific values — a grid earns its keep. Even then: gray, delicate, never dark. A delicate gray grid supports more accurate reconstruction of values than a dark grid that visually swamps the points being read.

Do / Don't: Grids

DoDon't
Delete the grid first; restore only if reading failsShip heavy or dark grid lines in any published graphic
Use a light gray grid only when look-up is the primary taskUse doubled grid lines or a box frame around panels
Use a white (negative-space) grid inside filled barsPlot on graph paper printed-side-up for publication
Keep grid stroke well under the data stroke weight (≈ 20–30% is a practitioner default — VDQI prescribes only muted and gray, no percentage)Let grid lines equal or exceed the data strokes
Prefer gray over thinned black when a grid is requiredPile a dense grid onto an already line-rich graphic

Failure mode — the grid that buries the data: a full dark grid over an age/sex pyramid or a trend line camouflages the very profile that matters (the notch, the staircase, the slope). Removing the grid lets the data silhouette speak.


§3. Ducks — Decoration That Becomes the Data

The duck is chartjunk at its most extreme: decoration that overwhelms, replaces, or structurally becomes the data. The name comes from the Big Duck, a duck-shaped retail building on Long Island (Flanders, NY) where the entire structure is its own sign — form swallows function. In Tufte's terms a graphic is a duck when decorative forms or computer debris take over, when the data measures and structures turn into Design Elements, and when the display purveys graphical style instead of quantitative information.

The architectural rule (Learning from Las Vegas)

Tufte borrows from Venturi, Scott Brown & Izenour: modernists who renounced applied ornament ended up designing buildings that were ornament. The maxim they quote — a warning from the nineteenth-century architect A. W. N. Pugin — transfers directly to data graphics (VDQI p. 117):

"It is all right to decorate construction but never construct decoration." — A. W. N. Pugin, quoted in Venturi, Scott Brown & Izenour, Learning from Las Vegas

Applied: you may style the axis labels, type, weights, and annotations (decorate construction), but the graphic's structure must never itself be ornament. When fake perspective, 3-D extrusion, or a pictorial frame is the structure, the graphic is a duck.

Named example: crosshatched bar chart → table

A common duck is a multi-category bar chart where each bar carries a different crosshatch (moiré on top of a duck) and the axis is choked with percentage ticks and all-caps labels. Tufte's remedy is to redraw it as a plain table. The table wins on every count: exact values instead of estimated bar heights, full category names instead of cramped labels, room for a second data column, and zero vibration. When the data is a handful of numbers attached to many words, the table is the graphic.

Interior decoration — symptoms

Treating a chart as a canvas to be styled rather than a message to be optimized. Watch for:

  • Fake perspective on bars or pies to look "modern"
  • 3-D extrusion that breaks comparison (the visible front face ≠ the value)
  • Pictorial fills (stacked coins for money, little people for population) that obscure the quantity they claim to show
  • Color gradients laid over areas whose value is already encoded by area
  • Drop shadows, glows, or embossing on chart elements
  • Complexity as a credential — "remarkable that the computer drew this" instead of "what interesting data"

Boutique data graphics — the high-fashion duck

Annual reports, mass media, and advertising actively cultivate the duck. Tufte's label is boutique data graphics: elaborate displays whose visual complexity is inversely related to their information content. Fake perspective is the signature move of the genre.

The compact diagnostic comes from Envisioning Information's diamonds/fishnet duck: "Everything counts, but nothing matters" (EI p. 34). In chartjunk every element clamors for attention and none carries meaning — the one-line test for any suspect display.

Production quality does not redeem a duck

Tufte concedes there are some superbly produced ducks — the California Water Atlas among them (p. 119) — but craft and junk are orthogonal: high production values decorate the failure, they do not cure it. At the other pole sits an American Education 3-D display spending five colors to carry five numbers, Tufte's candidate for the worst graphic ever to reach print (p. 118). Ducks also compound with vibration: a NEJM stacked pyramid's back planes optically flip forward (a Necker illusion) while its depth-stacked variable carries no label and no scale (p. 109).

Do / Don't: Ducks

DoDon't
Let the data's structure set the graphic's structureLet the graphic's structure turn into decoration
Use a table for a few numbers and many wordsUse a multicolor full-page illustration for ~5 data points
Design to provoke "what interesting data"Design to provoke "remarkable the computer drew that"
Decorate construction (style labels, type, weight)Construct decoration (extrude to 3-D, add fake perspective)
Keep every mark removable only at the cost of informationKeep any element that survives a "what does this encode?" test as "no data"

Failure mode — the sham dimension: adding a depth axis the data does not have. Extruding 2-D bars into 3-D both adds non-data ink and distorts the data ink, because the value now maps ambiguously to a front face, a back face, or a volume. This is the same offense the Lie Factor / dimensionality rule attacks (VDQI ch. 2): an n-dimensional quantity drawn in more than n dimensions overstates change. Cross-check sham dimension against tufte-graphical-integrity.


§4. Why Chartjunk Corrupts — and How to Detect It

Chartjunk is not neutral waste; it actively degrades reading through three channels:

  1. Perceptual masking. Vibration and heavy grids raise the visual noise floor, so genuine signals (the notch, the slope, the outlier) drop below it and become unreadable.
  2. False complexity. Crosshatch, borders, grids, and fills make a graphic look information-dense while it encodes five numbers; the reader spends attention decoding structure instead of reading data.
  3. Displaced trust. When the rendering shows off, attention shifts from message to medium — but a graphic's credibility rests on the data, not the sophistication of the draftsmanship.

Data-ink ratio as a chartjunk detector

Every chartjunk element lowers the data-ink ratio, so a chartjunk audit is a data-ink audit:

data-ink ratio = (ink used to present data) / (total ink used to print the graphic)
Chartjunk elementPenalty
Moiré / crosshatch fillAdds non-data ink
Heavy or doubled gridAdds non-data ink
Decorative border or frameAdds non-data ink
3-D extrusion / fake perspectiveAdds non-data ink and distorts data ink — doubly penalized

Any mark that cannot be defended as encoding data, or as a minimal aid (one axis line, a muted look-up grid), is chartjunk. See tufte-data-ink-ratio for the maximization procedure.


§5. Diagnosis Checklist

Run before publishing any chart. Any failed row names the chartjunk type to fix.

CheckPass conditionFail =
Are fills solid, gray, or open (no patterns)?YesVibration (crosshatch/moiré)
Does any series rely on hatching to be told apart?NoVibration
Is the grid lighter than the data strokes?Yes, or no gridGrid chartjunk
Is the grid gray rather than black, single not doubled?Yes, or no gridGrid chartjunk
Can you read the data profile with the grid hidden?YesGrid chartjunk
Does every remaining mark encode data or aid navigation?YesDuck (decoration)
Is the count of ink marks ~proportional to the data points?RoughlyDuck (false complexity)
More percentage ticks on the axis than data points?NoDuck (overcrowded axis)
Is the depiction flat — no perspective the data lacks?YesDuck (sham dimension / 3-D)
Would a plain table read more accurately than this chart?NoDuck (chart form unjustified)

§6. Remediation Patterns

Vibration fix

  1. Replace every crosshatch/pattern fill with solid fill, gray, or white.
  2. To separate multiple series, use a lightness ramp before reaching for any pattern.
  3. If the tool emits crosshatch by default, override it in the rendering layer — never accept default fills.

Grid fix

  1. Delete the grid. If the chart still reads, leave it deleted.
  2. If look-up is required, set the grid to ~10–20% gray at ~0.25pt against ~1pt data strokes (≈ 20–30% of the data weight). These numbers are practitioner defaults, not Tufte's — the book's rule is only that the grid be muted, gray, and visually subordinate.
  3. Alternative: a white grid — gaps inside filled bars, or grid lines erased to white over a tinted ground.
  4. Never plot on graph paper printed-side-up; use the unprinted reverse or plain paper.

Duck fix

  1. Remove everything that does not encode data.
  2. Flatten any 3-D extrusion to 2-D; delete fake perspective.
  3. If the structure itself is the decoration (the fill is the message), redraw from scratch as a flat chart — or a table.
  4. When the data is fewer than ~10 numbers and the story is comparative, ship a table, not a chart.

§7. The Bottom Line

No information, discovery, wonder, or substance is ever generated by chartjunk. The graphics that endure — Minard on Napoleon's march, Marey's train schedule, the newspaper weather history — are gripping because of narrative power, immense honest detail, and genuinely interesting data, never because of decoration.

"Forgo chartjunk, including moiré vibration, the grid, and the duck." — Tufte, The Visual Display of Quantitative Information


§8. Cross-book notes

The three redesign principles (BE p. 121) — the later books' cleanest statement of the chartjunk thesis, and the one to quote in a design argument:

  1. Clutter is a failure of design, not a property of the information. A crowded graphic does not prove the data was too much.
  2. Do not solve a visual problem by cutting content-resolution — that moves the display the wrong way.
  3. Fix the design instead. Tufte demonstrates this on a labelled scatterplot by pushing labels onto a separate visual level rather than deleting them.

Chartjunk of graph bureaucracy (BE p. 153) — a named second species alongside garish decoration, and the more common one in professional work: optically active grids, boxes and frames, redundant representations of the same datum, cross-hatched bars. The case study is the Boeing Columbia spreadsheet, where the most visually prominent things on the page are empty framing area and grid prisons around numbers too small to read. The tell is that the chart looks serious rather than gaudy — bureaucratic junk hides behind an air of rigor.

Frames as the loudest element (BE p. 62) — heavy frames activate the negative space between them, so the gutters out-shout the data. Tufte's audit question for any framed layout: do the most prominent visual effects on this page convey relevant content? If the answer is no, the frames are the chartjunk.

Visual clarity and intellectual clarity (VE pp. 47–49) — a rocket-silhouette bar chart of launch history is Tufte's specimen for decoration standing in for evidence, and it carries the argument that lack of visual clarity in presenting evidence tracks a lack of intellectual clarity about the reasoning. Junk is a symptom, not just a blemish. VE also supplies the institutional vocabulary — "administrative bloat" for the apparatus that accretes around evidence (pp. 118, 125, 149) — and the measurement to go with it: Reinhardt-style framing consuming 42% of a display's area (p. 118). Seeing With Fresh Eyes repeats the audit on an image matrix that turns out to be only 42% images, the other 58% frame furniture, oversized checkmarks, and undersized labels (p. 24). Measure the junk as a percentage of area; the number ends the argument faster than the adjective does.


Source: Edward R. Tufte, The Visual Display of Quantitative Information, ch. 5 "Chartjunk: Vibrations, Grids, and Ducks," with Pugin's architectural rule as quoted in Venturi, Scott Brown & Izenour, Learning from Las Vegas. Concepts paraphrased; quotations limited to single attributed sentences. Survey figures (journal moiré percentages pp. 110–112, the 18% doubled-grid cost p. 114) are verified against the 2nd ed. (2001); grid stroke-weight percentages are labeled practitioner defaults where they appear.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.