Unlikely Professionals

Design Principles

Data visualization standards after Edward Tufte

Every chart, dashboard, pipeline indicator, and data display across the Mostly platform follows these six principles. They are not independent—they reinforce each other. The goal is always the same: show the data with the least possible interference.

01 Brutal Simplicity of Thought and Clarity of Data
The Visual Display of Quantitative Information, 1983
“Above all else, show the data.”

Every element must serve the data. If it doesn't encode information, it is chartjunk—remove it. Clarity comes from elimination, not addition.

Do
  • Let data speak without decoration
  • Use white space as structure
  • Remove until nothing remains but data
Don't
  • 3D effects, shadows, bevels
  • Decorative gridlines or borders
  • Gradient fills, ornamental icons
02 Charting Change Across Multiple Variables
The Visual Display of Quantitative Information, 1983
“The best graphics show causality, mechanism, and explanation.”

The most powerful visualizations integrate multiple variables into one coherent view. Minard's 1869 map of Napoleon's Russian campaign encodes six variables—location, army size, direction, losses, temperature, and time—in a single image with no legend.

Encoding toolkit for Mostly: position (x/y) for two variables, color for a third (branch, status), size for a fourth (invoice amount), shape for a fifth (account type), border for a sixth (guardrail, RFI flag).

Do
  • Encode 4–6 dimensions in one view
  • Show mechanism, not just correlation
  • Let structure tell the story
Don't
  • Separate chart per variable
  • Legends requiring repeated reference
  • Color for aesthetics, not data
03 Small Multiples
Envisioning Information, 1990
“At the heart of quantitative reasoning is a single question: Compared to what?”

Identical charts, each showing different data slices, arranged in a grid with identical scales, axes, and sizes. The viewer learns one visual pattern and applies it across all panels. Small multiples make differences and patterns obvious without overloading a single view.

Do
  • Identical scales across all panels
  • Consistent size and shape
  • Order logically (time, magnitude)
  • Keep each panel ruthlessly simple
Don't
  • Vary axis scales between panels
  • Use for >20 categories
  • Overcomplicate individual panels

Mostly applications: Revenue by branch (5 identical line charts), project status distribution by branch, RFI aging by month, cert processing time by certifier.

04 Sparklines — Datawords
Beautiful Evidence, 2006
“Small, intense, simple, word-sized graphics with typographic resolution.”

Graphics should be as ubiquitous as text. No frames, axes, or tic marks—just the data trace, roughly 2cm wide, embedded inline with surrounding text, numbers, or table cells. Data-ink ratio = 1.0.

Types: line (trend), bar (magnitude), win/loss (binary), dot plot (distribution).

Mostly applications: Payment progress sparkline per invoice row, stage-progression sparkline per project, inline trend next to KPI numbers, RFI aging dot-plot per branch.

Do
  • Embed inline with text and tables
  • Keep word-sized (~2cm wide)
  • Zero non-data ink
Don't
  • Add axes or legends
  • Oversize beyond typographic scale
  • Disconnect from surrounding context
05 Escaping Flatland
Envisioning Information, 1990
“Escaping this flatland is the essential task of envisioning information.”

Represent 3+ dimensions on 2D surfaces without resorting to 3D charts. Tufte explicitly warns against them—3D pie charts distort slice sizes, 3D bar charts create false depth perception. Instead, use 2D encoding strategies.

The encoding toolkit:

Position (x/y) Dimensions 1–2
Color Dimension 3 — categorical grouping (branch, status)
Size Dimension 4 — bubble diameter, line width
Shape Dimension 5 — limit to 4–5 types
Texture Dimension 6 — solid, dashed, dotted
Small multiples Dimension 7+ — panel arrangement itself
Connection/flow Relational — lines encoding relationships
06 The Data-Ink Ratio
The Visual Display of Quantitative Information, 1983
“Erase non-data-ink, erase redundant data-ink—within reason.”

Data-Ink Ratio = ink devoted to data ÷ total ink. Maximize the proportion representing actual data. Every pixel earns its place or gets cut.

0.25
Poor — heavy decoration
0.55
Typical — chart tool defaults
0.80
Target — dashboards
1.00
Sparklines — zero non-data ink

Erasing strategies:

Gridlines Remove, or 1px light gray only where axis readings are ambiguous
Legends Label data directly instead of a separate legend box
Fills Flat color or transparent — no gradients, shadows, rounded corners
Axis labels Show 0, 50, 100 — not every increment
Pie charts Replace with horizontal bars
Color Use for data encoding only — never for decoration

“Within reason” — new audiences may need more annotation. Accessibility may require redundancy (labels + colors for color-blind users).

Our Visual Language

Tufte gives us the rules. These are the choices that make the work ours—drawn from visualizations we find stunning and want to emulate. Data should feel like a natural material, not a spreadsheet.

Minard, 1869
Six variables, no legend. The band is the army.
Pop/Rock Genealogy
Genres flow like strata—merge, split, fade. Labels embedded in the river.
Network Growth
Monochrome dots on cream. Density is the data. Ink = information.
University Sankey
Three views unified: scatter → summary → alluvial flow. One reading direction.
Computation Timeline
Bubbles sized by significance, connected by lineage. Five dimensions, zero clutter.
Earth-Tone Streamgraph
Data flows like geological strata. Warm monochromatic family. No legend needed.
Disk Space Radial
Overview radial + small multiple breakdowns. Metaphor earns the form.
Layered Data Cosmos
Multiple chart types coexist on one canvas. Color bleeding creates depth without 3D.

1. Cream, Not White

Every surface starts warm. #fffff8 is the floor, not #ffffff. Clinical white creates tension between the eye and the data. Warm off-white lets the ink settle into the page like letterpress on cotton stock. This applies to every background—dashboards, charts, panels, modals.

clinical
ours
warm
parchment

2. Data as Material

Data should have the visual weight of a physical substance—geological strata, flowing rivers, woven threads. Streamgraphs and alluvial diagrams give temporal data a presence that bar charts cannot. When data flows, merges, splits, and fades, the shape is the story.

Warm monochromatic strata—the palette of the earth-tone streamgraph.

Preferred forms: streamgraphs for temporal evolution, alluvial/Sankey for flow between categories, network graphs for relationships that grow over time, bubble timelines for significance + lineage. These forms make data feel alive. Bar charts and tables are not wrong—but when the data has movement, give it a form that moves.

3. Density as Beauty

Information density is not clutter. Clutter is decoration without data; density is data without decoration. The network growth chart (1982–2012) is packed with dots and connections—and it is beautiful because every mark encodes a real relationship. The pop/rock genealogy has thousands of labels embedded in flowing streams—and it rewards sustained attention.

Don't simplify by removing data. Simplify by removing everything that isn't data. A dashboard showing 200 projects with pipeline indicators, sparklines, and status dots is not overwhelming if every pixel earns its place.

4. Flow and Connection

When data has lineage, ancestry, or progression, show it with connections—not separate tables. Lines between nodes, bands between stages, threads between institutions. The university Sankey combines scatter plots, summary bars, and alluvial flow in one left-to-right reading. Three views, one story, one direction.

Flow forms
  • Sankey/alluvial for stage transitions
  • Streamgraphs for temporal evolution
  • Network graphs for relationships
  • Connected bubble timelines
When to use tables instead
  • Precise values matter more than shape
  • Data is categorical without movement
  • User needs to sort, filter, or search

5. Labels Live in the Data

Every reference work embeds labels directly in the visualization—genre names inside the stream, institution names at the end of the flow, year labels along the axis without a separate legend. When labels live in the data, the eye never leaves. When labels live in a separate box, the eye shuttles back and forth and the story breaks.

Streams Label text runs inside the stream band, following its flow
Sankey Destination labels at the right edge where the bands terminate
Timelines Year labels along the left margin, events labeled at the point
Networks No legend—density and weight are the label
Pipeline dots Status label centered under the current dot, not in a separate legend

6. Warm, Organic Palette

Our colors come from earth, not from a UI kit. Gold, amber, sienna, burgundy, warm browns, muted teal, dusty purple. Not neon. Not corporate blue. The streamgraph's monochromatic brown family, the university Sankey's muted jewel tones, the computation timeline's soft pastels on cream—these are our reference points.

Color should feel like pigment on paper, not pixels on a screen. When a chart uses only one hue family (browns, or ambers, or slate blues), the data reads as a unified material, not a collection of competing categories.

Platform Color Palette

The TUFTE_COLORS palette from the chart service. Muted earth tones—distinct but never loud.

gold
slate
purple
teal
red
lt blue
brown
forest
sienna
dk slate
dusty
mt teal
Design Workflow
Step 1

Start with brutal simplicity. What is the core story? Erase everything else.

Step 2

Measure data-ink ratio. Remove decoration. Target 0.7–0.85.

Step 3

Choose technique by variable count. 1–2 variables → sparklines. 3+ variables → small multiples (categories), multivariate display (relationships), or escaping flatland (dimension encoding).

Step 4

Test before shipping. Can anything be removed without losing information? If yes, remove it.

Pre-Ship Checklist
Before any UI, chart, or data display goes live
Source Books
1983 The Visual Display of Quantitative Information Principles 1, 2, 6
1990 Envisioning Information Principles 3, 5
1997 Visual Explanations Causality, mechanism
2006 Beautiful Evidence Principle 4 (sparklines)