How to stack multiple information streams on a flat surface without generating visual noise — the 1+1=3 principle, the noise formula, figure/ground, weight hierarchy, grids, type placement, and color as a layering tool, with worked editing techniques.
Installation
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
How to stack multiple information streams on a flat surface without generating visual noise — the 1+1=3 principle, the noise formula, figure/ground, weight hierarchy, grids, type placement, and color as a layering tool, with worked editing techniques.
Every display lives on flatland: a flat surface that must carry many overlapping streams of meaning. Layering is the technique for visually stratifying that surface — separating data from scaffolding, primary from secondary, read-now from consulted-occasionally — so the hierarchy of visual effects matches the hierarchy of information. Tufte's governing claim (Chapter 3 opener, p. 53): confusion and clutter are failures of design, not properties of the information. The fix is never to dumb the data down; it is to layer it. Effective layering is hard — for every excellent performance, a hundred clunky spectacles arise — and it gets harder on screens: the "video flatland" of computer terminals replaces the relative constancy of paper, so interacting layers in stacked windows, framed by system chrome, turn dynamically obtrusive in ways print does not.
§1. The 1+1=3 Principle
Place two marks on a surface and you always get a third element for free: the activated space between them. This is geometric and perceptual fact, not a stylistic choice. Two black lines on a page produce a bright white path that reads as its own visual entity whether you wanted it or not — and the path can even appear to have its own angled ends. Add more marks and the negative shapes proliferate exponentially.
Josef Albers proved the arithmetic with two cardboard strips ("One Plus One Equals Three or More," Search Versus Re-Search, 1969, cited p. 61):
Configuration
Result
1 strip + 1 strip, touching
2 widths (ground between them collapses)
1 strip + 1 strip, separated by 1 width
3 widths — the middle gap reads as a negative third strip
2 strips crossing at centers
4 arms, 4 rectangles, 4 triangles, 4 squares — many meanings from two marks
Crossing off-center
the in-between figures go unequal — even more shapes
Albers' move was to see area and surface rather than border and edge — which is how you escape the "magic" of optical illusion and reason instead about negative-space activation. Gaetano Kanizsa's subjective-contours demonstrations (Nolli/Turgot maps, illusory triangles; Scientific American, 1976, cited p. 60) make the same point from the other side: the eye constructs a whole shape — with a crisp edge — from marks that never draw that edge. Marks generate form in the gaps; the only question is whether that form is information or noise.
The two-step editing discipline (p. 61):
Find every 1+1=3 effect in the display — white paths, activated gaps, vibrating edges, halos.
Ask if each carries information. If not, it is noise. Mute or remove the mark that caused it.
Most of the time the surplus visual activity is non-information — noise and clutter. This two-step logic (recognize the 1+1=3 effects, judge whether they generate noise) is Tufte's master tool for refining designs, for graphical reasoning, and for the subtraction of weight.
Even the type around these examples interacts with them — the 1+1=3 reverberations are vivid precisely because everything is printed in strong black; high light/dark contrast accentuates the clutter (p. 61, fn. 8). The interaction effect is unavoidable, so manage it.
§2. The Noise Formula
Tufte states the relationship as a law, not a preference (p. 62):
"The noise of 1 + 1 = 3 is directly proportional to the contrast in value (light/dark) between figure and ground." — Tufte, Envisioning Information, p. 62
Implications:
On a white page, lighter marks generate less incidental clutter than dark ones. A gray line and a black line at identical weight do not produce identical noise — the gray one produces far less. This is perceptual arithmetic, not taste.
Therefore prefer a varying range of lighter colors for secondary layers on white grounds.
Noise reduction is not cosmetic. Signal enhancement reduces viewer fatigue and improves reading accuracy at a computer interface, a flight-control display, or a medical instrument. Clarity is not everything, but there is little without it.
The worked technique: thin two sides of every block
Tufte's three-map sequence (Jon Wertheimer, Yale student project, p. 62) shows the formula applied as a concrete editing move, not just a principle:
Map
Treatment
Effect
Top (poor)
every block bordered by bold lines of equal weight
bold shapes vibrate all over; even with only nameless streets the map is already in visual trouble
Center (better)
thin two sides of each block, so every street is bordered by one thick + one thin line
the thin lines read like gray lines — visually light — which deflects the 1+1=3 effect at each street
Bottom (best)
gray edges throughout
serene, motionless edges that can absorb additional geographic detail without new noise
The move: do not lighten everything uniformly. Pair a heavier line with a lighter one along each boundary so the boundary stops vibrating, while still reading as an edge.
Failure case: the worst index ever designed (p. 62)
A flying-aircraft instruction manual whose index — meant for use during actual flight instruction — was set so that 1+1=3 vibration shimmers across the entry list, and the index has no page numbers at all. Tufte's "rare perfect failure": maximum visual noise plus zero retrieval information, in a document you read while flying a plane. The lesson: noise is most dangerous exactly where reading is hardest.
§3. Figure and Ground
Information consists of differences that make a difference (p. 65). Every surface splits into figure (read as data) and ground (what it sits against); layering is how you enforce the meaningful differences. When layers are not differentiated, the unrelenting interactive arithmetic of flatland takes over and 1+1=3 or more.
The undifferentiated-surface test (Simla, India vs Tokyo, p. 58)
Walk the display and ask: do contours, roads, rivers, names, and annotations all sit at the same value, texture, color, and near-same shape? If yes, the surface is undifferentiated. The U.S. Army map of Simla is Tufte's specimen: every element at one visual level, producing a surface he calls jumbled up, blurry, incoherent, chaotic — unintentional optical art, "a failure to communicate" (p. 58).
The Tokyo Prefecture map (1884, p. 58) is the corrective: it separates layers by distinctions in shape, value (light to dark), size, and especially color; the negative areas carry information too (light strips formed by the building grid read as roads and paths); the water symbol is a blue field that fades at each outlined edge; and the whole sits on a dull background rather than bright white, so the colors stay calm and distinct instead of clamoring.
Figure/ground failure modes
Failure mode
Mechanism
Specimen
Undifferentiated surface
all elements at equal value/texture/color/shape
Simla, India (U.S. Army map U 502)
Ground louder than figure
scaffolding outweighs data
ECG trace caught in a thick grid; NJ Transit grid fighting the times
Collapsing layers
two layers lose separation when contrast drops
Paul Klee Symptomatic sketch: red commentary and black line merge once light/dark and color differences are minimized (p. 55)
Activated negative space
marks generate unintended shapes between them
word boxed in a border (halos at every corner); striped index
Moire / vibration
two regular patterns near the same frequency beat against each other
Nolli 1748 Rome: heavy river inking makes bridge names and a boat vibrate in a "moire prison" (p. 60)
§4. Color as a Layering Tool
Color is the most powerful layering device on flatland: it separates streams without adding a single new mark. Used wrong, it collapses layers or manufactures new noise.
Imhof's two rules (Eduard Imhof, Cartographic Relief Presentation, p. 72, quoted by Tufte twice)
Large areas, muted (p. 58, fn. 3 — Imhof's "first rule of color composition"). Pure, bright, or very strong colors have loud, unbearable effects when spread unrelieved over large adjacent areas; the same colors achieve extraordinary effect used sparingly on or between dull background tones. Imhof quotes Windisch: noise is not music — only on a quiet background can a colorful theme be built. The background's quietness is what gives accents their force.
Small areas, intense (p. 63, fn. 10 — the complementary rule). Confine strong, heavy, rich, solid color to the small areas of extremes; let large background and base areas work quietly — muted, grayish, or neutral — so the small bright areas stand out most vividly. Tufte calls small spots of intense saturated color carrying information a design secret of classical cartography (and of traffic lights).
Worked color-layer examples
Uboku Nishitani calligraphy (p. 53): a second color (red commentary) annotates black brush strokes. By forming a distinct layer the red keeps detail, coherence, and serenity while sitting crisply beside the gestural black line. Its saturation partially offsets its lighter value and finer line — appropriate to meticulous annotation. Alone each color is a strong statement; together, stronger.
IBM Series III copier exploded diagram (p. 54): color effortlessly differentiates annotation from annotated, letting 300 small parts and their identifying numbers coexist without collision. Color carries the layer that would otherwise need extra marks or callout boxes.
a saturated blue or red covering most of the chart
Small areas intense
one saturated spot marks the data extreme
uniform medium saturation everywhere
Saturation encodes rank
most important layer = highest saturation
equal saturation = no hierarchy
Value encodes depth
data dark, scaffolding light
grid as dark as the data
Color separates, doesn't shout
color the lights so each signal reads as its own layer
strong shapes + strong color + grid all at one nervous level
§5. Weight Hierarchy and Grids
Weight hierarchy assigns line weight, type size, and mark intensity in proportion to information importance. Scaffolding — grids, rules, borders — gets the least. The decisive quality is relational: each layer's visual weight must be in proportion and in harmony with the substance it conveys (p. 54). Concrete test: if the grid is more prominent than the data, the proportion is wrong.
Tschichold's table-rule hierarchy (Jan Tschichold, Asymmetric Typography, 1935/1967, p. 62, cited p. 55)
Treatment
When to use
Tufte/Tschichold rule
No rules
default — try this first
"first, try to do without rules altogether"
Thin horizontal rules only
columns wide enough to read without guides
none beats thin, thin beats thick
Thin vertical rules
only when columns are so narrow that reading errors occur
needed only then
Screened / gray grid
reference grid behind dense data (ECG leads, music staves)
gray almost always works; a delicate gray line can beat a heavy grid for accuracy
Black / heavy grid
never, for data scaffolding
see the quote below
"Dark grid lines are chartjunk." — Tufte, Envisioning Information, p. 59
"Tables should not be set to look like nets with every number enclosed." — Jan Tschichold, quoted by Tufte, p. 55
Implementation note (not Tufte's number): Tufte gives no percentage or point size for "gray" or "thin" — he specifies only gray, screened, thin, muted relative to the data. The book's actual examples are the Stravinsky Sacre du printemps sketchbook grid (a gray grid that quietly and precisely locates the music) and the screened ECG grid behind 12 monitoring leads. If you must pick numbers, treat a light-gray reference grid against full-strength data as a sensible default — but flag it as a practitioner convention, not a value Tufte states.
Grid escalation rule
A grid earns darkening only when the graphic is a genuine look-up table requiring interpolation — which Tufte notes is rare. Even then, keep it muted relative to the data. Practical paper rule (p. 59): with ready-made graph paper printed dark, use the reverse imprinted side so lines show faintly; if it is heavily gridded on both sides, throw it out.
The self-important grid (Dioscorides birds, p. 64; Tukey multi-window plot, p. 64)
Decorative gridding is an epidemic of data-imprisonment: stylized grids, little boxes around words, cadenced accents — all empty of content, present only so the work looks "designed." In the Tukey multi-window statistical plot, the most visually active elements are glowing optical white dots at every grid intersection, and the doubled, tremor-inducing lines consume 18 percent of the chart; redrawing in "ungrid style" removes the noise and returns attention to the data. Tufte's verdict on the ornamental grid (illustrated by the 6th-century Dioscorides bird illumination): it is good only as a perch for birds — it carries no information.
Leader dots as separators (NJ Transit, p. 55)
Tiny leader dots between rows read as gray: they make a distinction without erecting a barricade. They are small enough that they do not generate competing 1+1=3 paths — a separator with almost no noise cost.
§6. Subtraction of Weight
Layering is usually achieved by removing weight, not adding it (p. 60). Small, modest design moves can yield decisive results. Tufte ties this to Italo Calvino's literary method (Six Memos for the Next Millennium, pp. 3–4, cited p. 60): a working method of subtracting weight — from people, cities, the structure of stories, language itself — against the inertia and opacity that stick to creative work unless one finds a way to evade them.
Do / Don't
Do
Don't
erase marks that carry no information (the screened ECG grid removed nothing informative — only the noise hiding the signal)
add a box, rule, or fill to "organize" a layer
mute an over-inked element so the next layer comes forward (Nolli: muting the river calms vibration, brings bridge names forward, and keeps the rippling-water symbolism)
brighten or thicken to assert importance
extend the value range — push background lighter, data darker — to separate layers without moving them
leave competing layers at similar value and hope position alone separates them
Worked subtraction: the marshalling-signals redesign (p. 63)
The original aircraft-marshalling chart is an "information prison": a narrow range of strong shapes where grid, silhouette, and type all compete at one nervous visual level — too loud and too similar — and thick grid-box bars spawn 1+1=3 paths around every figure. The revision:
Extends light-to-dark range — gray calms the contrasty silhouette, which in turn lets the lamps and their motion read.
Colors the lights — small saturated spots separate the signals from everything else.
Erases ~460 lamp-whiskers that had read as glowing light and trembling motion (pure 1+1=3 noise).
Differentiates the type — set in Gill Sans, no longer equal in weight to the motion arrows; the title emerges from its "foggy closet."
Result: figure separates from ground cleanly, with strictly less ink.
§7. Type on a Layered Surface
Type is itself a layer, and it interacts with the marks under it.
Box-around-text noise (p. 62)
Surrounding words with a little box activates the negative white space between word and box (Tufte's specimen: a boxed Surgeon General's warning). Unless deliberate obscurity is the goal, do not box running words — use proximity, indentation, or weight instead.
Name-on-graphic placement: the Rhein sequence (p. 62, Imhof fn. 9)
Tufte's poor/poor/poor/good sequence of a river labeled "Rhein": the first three placements set the name badly, with an awkward white stripe materializing between the name and the river line.
Placement
Why it fails / works
name crowded beside / below the line
a 1+1=3 white stripe forms between type and graphic
name riding on ascenders (map 3)
a diverting white shape is formed by the ascending letters
name set from above the line (good)
type from above adjusts to graphics better — most words have fewer descenders than ascenders, so the underside of the word is smoother and sits cleanly against the graphic
Rule: place labels so the smoother edge of the word (the side with fewer ascenders/descenders) faces the graphic, and watch for the white stripe between type and line. These small local details accumulate across the whole surface and decide quality.
§8. Worked Layering Examples
Example
Layers stacked
What made it work
NJ Transit timetable (pp. 54–55)
times, stations, train IDs, grid
original put the least important datum (a 4-digit train ID used only by railroad staff) in the top row in the largest type, under a heavy stripy grid; redesign moved New York departure times to the very top, de-emphasized minor data, and replaced the grid with an implicit one defined by gray leader dots
Mrs. K hospital costs (p. 56)
time sequence + accounting line items + clinical commentary over 26 ICU days
three simultaneous voices each kept to a distinct register (dense billing column, larger narrative in the margins, dates as the chronological anchor) — a layered polyphony where no voice fights another
ECG / 12-lead trace (p. 59)
signal trace + reference grid
thick grid swallows the trace; the screened-down gray grid stays behind all 12 leads, calibrating without competing
Nolli map of Rome (p. 60)
streets/buildings + river + names
mute the over-inked river to stop the moire prison; names and boat come forward
§9. The Eastern Tradition of Figure/Ground
These are not data-visualization inventions; they are the oldest discipline of mark-making (p. 65). Tufte cites Gyorgy Kepes (The Language of Vision, 1948) and Lao Tzu: a vessel is useful through its hollow interior, a window through the space it opens — the non-existent element makes the thing serviceable. Applied to marks: a character's legibility depends on the distinctness and variety of the surrounding ground; the greater the variety and distinction among the background units, the clearer the character. Chinese and Japanese painters leave large empty paths across the surface, dividing it into unequal intervals so the eye moves at varying velocity and is forced to track relationships; the blank areas get as much consideration as the strokes. Calder's pen separates a stag from its watery reflection with the most economical of means. Blank space is a layer — design it, do not just leave it.
§10. Named Failure Modes
Failure mode
Mechanism
Signal
Fix
Data imprisonment
grid boxes enclose every cell
grid louder than data
remove vertical rules; thin horizontals only; often no rules at all
Equal-weight surface
all marks at one value/size/color/shape
jumbled, blurry, incoherent (Simla)
assign weight in proportion to information rank
Vibrating edges
heavy figure against white ground
edges shimmer or pulse
lighten the figure, or pair thick + thin along each edge, or mute the ground
Moire prison
two regular patterns near one frequency
optical beating
mute one pattern; screen the grid; avoid two overlapping regular grids
Information prison
strong shapes + grid + type at one nervous level
decoration becomes the main statement
extend value range; erase containers; differentiate the type
Ground louder than figure
scaffolding heavier than data
viewer reads the grid, not the data
screen the grid gray; data at full strength
Proportionality inversion
least-important datum in the most prominent slot
layout hierarchy contradicts information hierarchy
reorder by importance; top/largest = what the reader needs first
Box-around-text
words enclosed in borders
white halos flicker at the corners
drop the box; use proximity, indentation, or weight
Label white-stripe
name placed so a gap forms against the graphic
bright stripe between type and line
place from above; face the smoother edge of the word at the graphic
Decorative gridding
grids/boxes/accents added to look "designed"
active marks are empty of content
redraw "ungrid"; the self-important grid carries nothing
§11. Layering Audit Checklist
Name every layer: data, grid, labels, annotations, borders, background, blank space.
Rank by information priority: which layer must be found first? which is consulted occasionally?
Match weight to rank: does the most important layer have the most visual weight, the grid the least?
Find every 1+1=3 effect: walk each pair of adjacent marks; name the negative shape; classify it information or noise.
Apply the noise formula: is the value contrast between secondary layers and ground higher than it needs to be? Lighten what you can.
Check the grid: gray/screened? If black, it is chartjunk — darken only for a true interpolation look-up table, and even then keep it muted.
Check color: strong saturation confined to small areas of the top layer? Large areas muted, background dull rather than bright white?
Check for imprisonment: is anything boxed (cells, words, signals)? Remove the box first, then test whether any rule is needed.
Check labels: any white stripe between type and graphic? Re-place from above; face the word's smoother edge at the line.
Extend the value range: background lighter, data darker — is the separation now obvious without repositioning anything?
Apply Calvino's test: what can be subtracted without losing information? Subtract it, then re-check the surface.
Design the blank space: are the negative areas doing work, or just left over?
Source: Edward Tufte, Envisioning Information, Chapter 3 "Layering and Separation," pp. 53–65.
