agentleFS
jpoindexterjpoindexter/tufte-skills/skills/tufte-data-maps/SKILL.md

tufte-data-maps

Skill1 starsChanged 4 months agoMITView on GitHub

Use when designing or auditing a map that carries data and something is off — a plan view spending both axes on geography the question does not need, poster-thin density that collapses on close inspection, ornament filling empty regions, one rigid scale forcing a choice between detail and overview, aggregation to political areas distorting the signal, or map small multiples restyled per panel.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 6 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    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:

    Every information surface — paper, screen — is two-dimensional, but data and nature are multidimensional.

    line 36
  2. Distinctive:

    Use a perspective/pictorial register where local texture matters.

    line 81
  3. Distinctive:

    When variables are interrelated, let a single graphical mark carry several readings at once instead of spending one mark per variable.

    line 106
  4. Distinctive:

    DON'T lock the map to a single scale: when that forces you to drop either the detail or the frame.

    line 149
  5. Distinctive:

    DON'T spend a separate encoding channel per variable: when the variables co-vary or share an axis.

    line 158
  6. Distinctive:

    State the data question first: Write the finding the map must deliver.

    line 172

≈ “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

Tufte: Data Maps as the Standard of Information Design

Overview

In Envisioning Information Tufte does not treat the map as one chart type among many; he treats high-quality maps as the gold standard of all information design — bountiful detail, several layers of close reading combined with an overview, rigorous survey data. A good map is the most demanding escape from "flatland": it increases (1) the number of dimensions shown on a plane surface and (2) the data density per unit area, and it rewards both a distant macro glance and close micro inspection. The chronic failure is the "duck" — pretend dimensions, ornament, and posterized thinness substituted for data.

Scope note. This skill covers Tufte's map argument in Envisioning Information, chiefly ch. 1–2 (Escaping Flatland, Micro/Macro Readings). EI does carry its own choropleth critique: pp. 40–41 fault the shaded "blot map" on three grounds — non-uniform area sizes, color filling often-nearly-empty land so large unpopulated regions get the greatest visual emphasis, and historical boundary changes breaking statistical comparison — and present the fix, mesh maps built on Japan's division into 379,000 equal-sized cells (developed in tufte-micro-macro-readings; cite EI pp. 40–41 for it). What remains VDQI territory: the fuller "area is not data" patch-map treatment and classification schemes. Modern conventions (Jenks, Moran's I, ColorBrewer) are out of scope for both sources.


§1. Maps set the standard; posters do not

Tufte's central claim for cartography: excellence in information design is calibrated against good maps, and most failed graphics fail because they behave like posters instead. Posters are made for viewing across a room — strong images, large type, thin data. Maps are made for sustained close reading — dense data, layered detail, an overview that survives inspection. (EI, p. 35)

"It is very much like an excellent map, but with many dimensions breaking free of direct analogy to conventional cartographic flatland." — Tufte, Envisioning Information, p. 26

The test: does the display reward both a distant glance and a nose-to-the-page inspection? If it only works from across the room, it is a poster, and the data is too thin.

Poster mindset (avoid)Map mindset (aim for)
Read from a distance onlyRewards distant macro view and close micro reading
Strong single image, large typeDiverse bountiful detail, many layers
Thin data densityHigh data density per unit area
Ornament enlivens "boring" numbersData is the interest; design is self-effacing
One reading, one paceReader chooses pace, path, and depth

§2. Escaping flatland: increase dimensions and increase density

Every information surface — paper, screen — is two-dimensional, but data and nature are multidimensional. Tufte's metaphor: like a toad's shed skin collapsing out of "spaceland" into flatland, our displays flatten richly dimensional information (EI, p. 14). The two design levers that fight this (EI, p. 13):

  1. Dimensionality — the number of variables you can register on a plane surface.
  2. Data density — the amount of information per unit area.

Operating rule, from the Java timetable: in flatland every already-available axis must carry data — no dimension should be left empty or spent on ornament (EI, p. 24). Use the margins, use the vertical, use the slope of a line, use color.

Density benchmarks Tufte cites (real numbers, not estimates):

Map / displaySourceAchieved density
Constantine Anderson axonometric map of midtown ManhattanEI p. 371,686 building/store/park names + 657 street addresses on one 60×92 cm (24×36 in) sheet; ≈3 characters per cm² (20 per in²)
LA air-pollution relief mapsEI pp. 28–292,400-cell spatial grid (5 km cells) × 12 maps = 28,800 readings; 5 variables total
Java graphic railroad timetableEI pp. 24–266 variables encoded in each train diagonal; 16-variable internal schedule
Modern Maunder "butterfly" sun mapEI p. 23≈10× the data density of the 1904 original; ~a century of data, nine solar cycles
Aerial photograph (Senlis, France)EI p. 3810⁶–10⁸ bits required just to digitize the detail

§3. The data question governs the projection, not cartographic convention

Tufte's strongest map lesson here: choose orientation and projection to free a dimension for the data variable, even if that abandons the conventional plan view.

  • Profile / section over plan view. A Japanese newspaper weather map shows a side profile of Japan, an "ocean-eye view," with gray contours tracing the 0 °C and −10 °C surfaces through the clouds. The vertical axis carries temperature and altitude — exactly what a traditional plan-view weather map throws away by spending both its axes on latitude and longitude. (EI, p. 28) The trade-off: a profile works best for long, thin countries.
  • Linear-distance maps. A graphic timetable collapses the three spatial dimensions of travel into one track-relevant dimension — distance along the rails — when only along-track position matters. Station rows are spaced in proportion to real along-track distance, so a constant-speed train draws a straight diagonal. (EI, pp. 24–26)
Data questionMap form to reach forEI exemplarWhy it wins
Where does each value sit in 2-D space, plus one or more extra variables?Relief/surface small multiplesLA smog (pp. 28–29)each panel a surface for a 3rd variable; panels form a matrix
How does a quantity vary along a single path (route, track, river)?Graphic timetable / linear-distance mapJava railroad (pp. 24–26)collapse 3 spatial dims to track-distance; free other axes for time, speed
Does the traveler only need the route itself?Strip / itinerary mapLondon–Dover / Calais–Paris (pp. 112–113)trades flatland for "lineland" — context is sacrificed, but the viewer can edit down to and focus directly on the relevant
Does the answer live in a vertical/sectional dimension the plan view hides?Profile / section mapJapan weather (p. 28)side view frees the vertical for temperature/altitude
Need fine local detail and a global frame at once?Multi-scale / axonometric mapIse Shrine (p. 13), Manhattan (p. 37)combines a close view with an overview
Must every mark carry several variables at once?Multivariate tinted mapGEBCO bathymetric chart (p. 91)each color mark signals four variables — latitude, longitude, sea-or-land, and metric depth or altitude
Are many layers jumbling into one visual level?Layered map — separate by shape, value, size, colorSimla vs. Tokyo Prefecture (p. 58)the Simla map's equal-everything jumble vs. Tokyo's calm layered separation; mechanics in tufte-layering-and-separation
Is the lat/long plan view itself the literal answer?Conventional plan-view map(traditional weather map, named as the foil)only when both 2-D axes genuinely carry the question

Strip maps, expanded (EI pp. 112–113). The 1801 London–Dover / Calais–Paris guide compresses each leg into a narrow vertical strip: a deliberate "wrong-way" departure from flatland toward lineland whose merit depends entirely on what is done with the freed second dimension. Two design details: the normally fixed compass pointer rotates strip by strip as the route bends, re-referencing orientation to the traveler rather than to north; and 19th-century Japanese book-length tour guides go further, printing the journey as one continuous horizontal strip gathered into bound, flowing pages — no strip discontinuities — with the vertical writing system conveniently labeling the horizontal path (as in Moronobu's 1690 Tokaido panorama).


§4. Multi-scale maps: marry local detail to a global overview

Don't lock a map to a single scale. The Ise Shrine travel guide shifts deliberately from a friendly bird's-eye perspective (local detail around the shrine) to a hard flat railroad map on the right margin (the national network linking the shrine to major cities). "A change in design accommodates a change in the scale of the map," giving the reader both a place of detailed refuge and a broad overview in one sheet. (EI, p. 13)

  • Use a perspective/pictorial register where local texture matters; switch to a flat schematic register where structural overview matters.
  • Make the registers point at each other — the Ise guide's stand-up labels point precisely to each location, stitching the two scales together.
  • A single rigid scale forces you to either drop local detail or drop the overview. The hybrid keeps both.

§5. Micro/macro reading: to clarify, add detail

Tufte's most counterintuitive map principle. Dense maps are read at two scales at once: micro (personal close-up stories in the data) and macro (overall pattern). The Constantine Anderson axonometric Manhattan map renders individual windows, subway entrances, bus shelters, telephone booths, canopies, trees, and sidewalk planters; its thousands of tiny windows, seen from a distance, gray into surfaces that resolve into whole buildings. Detail cumulates into coherent macro structure rather than fighting it. (EI, p. 37)

"Simplicity of reading derives from the context of detailed and complex information, properly arranged." — Tufte, Envisioning Information, p. 37

"to clarify, add detail." — Tufte, Envisioning Information, p. 37

Reading scaleWhat the reader getsDesign obligation
Microindividual stories, a credible refuge where the pace of viewing slows and personalizesrender real fine-grained detail; don't summarize it away
Macrooverall pattern, freedom to compare and sort, an overviewarrange detail so it aggregates into coherent larger structure

The Senlis aerial photograph makes the same point: micro-details (houses replacing the old Gallo-Roman fortification ring) mix into an overall town-plan pattern visible only at the macro scale (EI, p. 38). Practical inversion: when a map feels cluttered, the fix is usually better arrangement of more detail, not less data. Stripping data to "simplify" produces a poster.


§6. Multifunctioning marks: one element, many variables

When variables are interrelated, let a single graphical mark carry several readings at once instead of spending one mark per variable. The Java timetable's train diagonals each encode six variables simultaneously (EI, p. 26):

#VariableHow the diagonal encodes it
1Location of the train between townsvertical position against the station rows
2Time of that positionhorizontal position against the 24-hour axis
3Direction of travelsign of the slope (down-right vs. up-right)
4Train typeline style, from a 2-D type × seasonality matrix
5Relative speedsteepness of the slope (steeper = faster)
6Yearly/seasonal pattern of operationthe same type × seasonality matrix

A separate 16-variable schedule rides alongside. (Tufte develops "multifunctioning graphical elements" formally in VDQI ch. 6; here the map is the demonstration.) Rule: if two variables co-vary or share an axis, look for one mark that expresses both before you add a second encoding channel.


§7. Maps as small multiples and as entries in a larger matrix

A data map need not stand alone. The LA air-pollution display is a small multiple: twelve maps, each a relief surface of one pollutant over the LA basin, with an identical design repeated across panels. The repetition buys an "economy of perception" — decode the layout once, then read every panel for free. Each small map reports the 2-space location of a third quantity, and the maps themselves become cells in a time-of-day × pollutant matrix, for five variables in all. (EI, pp. 28–29)

  • The children's-shirt color array makes the underlying mechanism explicit: multiplied smallness forces comparison within the eyespan, so the active eye contrasts adjacent panels instead of relying on memory of images scattered across pages (EI, p. 33).
  • Constancy is the whole point: hold scale, projection, framing, and color mapping identical across panels. A steady canvas makes the data the only thing that changes between maps. Restyle a panel and the eye reads the chrome difference, not the data difference.

§8. Named failure modes

Failure modeWhat it isSymptomFix
The duck (false escape from flatland)a pretend extra dimension or ornament bolted onto a thin data seta content-empty 3rd dimension; the royal-dining-table woodcut whose bad drawing won't "hold" the pots; the diamonds-on-fishnet thigh-graph (EI pp. 34–35)strip the fake dimension; spend that surface on real data
Map-as-poster (data posterization)designed for distant viewing with strong images, large type, thin densityreads fine from across the room, collapses on close inspectionraise data density; add layered detail that rewards close reading
Swift's elephants (chartjunk filling empty space)decorating blank map regions with ornament instead of dataheraldic rosettes/symbols planted where there are no spots (Scheiner's sun maps); "elephants for want of towns" (EI p. 21)leave empty space empty, or fill it with data — never ornament
Plan-view wastespending both map axes on latitude/longitude when the data question needs a freed dimensionconventional weather map can't show the vertical temperature structurere-orient or re-project (profile/section) to free an axis for the variable (EI p. 28)
Relief masking3-D surfaces over a map base hide low values behind tall foreground peaks"except for those masked by peaks" — occluded readings (EI p. 29)rotate/order surfaces, use transparency, or break into small multiples so nothing is occluded
Marginal-only displayshowing interrelated variables one-at-a-time-in-parallelmargins shown, joint structure hidden (butterfly-diagram caution, EI p. 23)show the joint distribution, not just the separate margins

§9. Do / Don't pairs

DO let the data question pick the projection — a side-profile of Japan to carry temperature in the vertical. DON'T default to the lat/long plan view out of cartographic habit when it spends both axes on geography you don't need.

DO combine a detailed local view with a global overview — the Ise Shrine bird's-eye plus national railroad map. DON'T lock the map to a single scale when that forces you to drop either the detail or the frame.

DO add structured detail to clarify — the Manhattan map's windows that gray into buildings. DON'T strip data to "simplify" — thinning produces a poster, not a clearer map.

DO keep panel design identical across small-multiple maps — the repeated LA-smog layout. DON'T restyle each panel — the eye should compare data between maps, not chrome.

DO let one mark carry several interrelated variables — the Java train diagonal's six readings. DON'T spend a separate encoding channel per variable when the variables co-vary or share an axis.

DO use every available dimension to carry data — the timetable's margins, profile, and slope. DON'T leave an axis empty or fill it with ornament — Swift's elephants in the blank spaces.

DO make the design self-effacing so focus falls on data, not the data-container. DON'T add pretend dimensions or "duck" decoration to dress up a thin data set.


§10. Application checklist

Apply in order when building or auditing a data map:

  1. State the data question first. Write the finding the map must deliver; let it pick the form from the §3 table before you pick a projection.
  2. Audit every axis for data. Vertical, horizontal, slope, color, margin — is each one carrying a variable, or sitting empty/ornamental? (§2)
  3. Check both reading scales. Does it reward a distant macro glance and close micro inspection? If only distant, it's a poster — add detail (§1, §5).
  4. Look for multifunctioning marks. Are co-varying variables collapsed into shared marks, or wastefully split across channels? (§6)
  5. If repeated, lock the canvas. For small-multiple maps, hold scale/projection/framing/color constant so only data changes (§7).
  6. Hunt the failure modes. Ducks, posterization, ornament in empty space, plan-view waste, relief masking, marginal-only display (§8).
  7. Confirm self-effacement. The reader's attention should land on the data, not on the cleverness of the graphic.

Cross-book notes

  • VDQI pp. 16–27 is the pack's richest data-map vein: the cancer atlas (3,056 counties, ~21,000 numbers per map), Snow's cholera dot map, the French wine-exports flow map, the 1.3-million-galaxy map (2,275,328 rectangles in ten grays), and the observation that data maps arrived only in the 17th–18th centuries despite millennia of cartography — cite VDQI, not EI, for these.
  • Beautiful Evidence: Swiss topographic maps as the standard for contour lines flowing over place-names (pp. 58–59); cartographic legends distinguishing 34, 15, and 17 line types with typographic clarity (p. 71); and the blunt rule "Maps don't put boxes around city names" (p. 79).
  • Visual Explanations: the Chinese Yu ji tu grid map of 1137 — a coordinate grid of 100-li squares, working out to roughly 1:4,500,000 — is the pack's earliest data-map exemplar (p. 14; VDQI pp. 20–21 carries the same map). And Matthew Paris's map of Britain, c. 1250, annotates its own distortion on the sheet: the island should have been drawn longer, had the page allowed it (p. 24). A map that names the compromise its medium forced on it is the honest-scaling precedent — declare the distortion rather than let the reader discover it.
  • Seeing With Fresh Eyes: Swiss topographic sheets are the master case — place-name type scaled to settlement population, index contours, spot heights, even lake-bottom depths, with rock rendering pitched differently at 1:25,000 and 1:50,000 (pp. 9, 32, 62). Tufte's summary claim: cartography has already solved the display problems the rest of the field is still arguing about — colour, typography located in content, massive data, and readers of wildly different expertise. Its own limit is worth carrying too: an aerial view is no help to someone standing on the ground, so the map model breaks wherever the reader's viewpoint is inside the terrain rather than above it.

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.