Two datasets per shape
Every graph we teach gets two, from two different domains. One that affects them, one Earth and space. Operator, 2026-08-27: "I need two different types of data for each graph we are teaching."
Why two
A student who only ever meets a choropleth of pollution burden learns that a choropleth means pollution. The shape and the subject fuse, and the shape stops transferring.
Two datasets in two domains break that fusion in the cheapest possible way — same shape, same period, nothing in common but the geometry. It is also how the room gets co-taught: the left column is the civic lane, the right column is the science lane, and the shape is the thing both teachers are actually teaching.
The third column is the point of the pair. A pair earns its slot when the two datasets do not behave the same way in the same shape.
The pairs
| Shape | Affects them · LA | Earth & space | What the PAIR teaches that neither does alone |
|---|---|---|---|
| the table | their own collected data — lunch, attendance | USGS raw earthquake rows, last 30 days, SoCal | forty rows of theirs feels legible; four thousand of the quakes does not. The table fails at a size you can feel. |
| the bar | 311 requests by council district | earthquakes by magnitude band | one is a ranking of places, one is a ranking of sizes — the bar does not care, and the reader must |
| the number | today's AQI at the nearest monitor | current CO2 ppm, Mauna Loa | 42 and 420 mean nothing without a comparison, and neither number tells you which way is bad |
| the line | LA average summer temperature by year | Mauna Loa CO2 since 1958 | one is noisy and one is the smoothest line in science. Same shape, and only one of them lets you see a trend by eye. |
| the donut | where the school day goes, by minutes | composition of the atmosphere | theirs has 6 slices, the atmosphere has 2 plus a rounding error. A donut cannot show you the 0.04% that runs the climate. |
| the picture graph | street trees by species near campus | planets by number of moons | icons work for 40 trees; Saturn has 140+ moons. The icon is a unit, and it breaks at scale. |
| the histogram | 311 requests per day (Census commute needs an API key) | earthquake magnitudes, SoCal | BUILT + measured: they skew in OPPOSITE directions — 311 left (a tail of quiet days), quakes right (a tail of big ones). Quake counts fall 1684/721/88/14/1: Gutenberg-Richter in their own file |
| the scatter | tree canopy vs surface temperature by tract — NOT BUILT, needs a tract join | star temperature vs luminosity — the HR diagram | one shows a trend, the other shows clusters and a gap. A scatter is not a device for finding lines. |
| the dotplot | hours of sleep, class-collected | distances to the nearest 50 stars | both keep every value; one is 30 dots, the other is wildly uneven spacing |
| the box plot | household income by council district | monthly temperature at one station | the mean hides, twice, for opposite reasons — income is skewed by the top, temperature is bimodal by season |
| the choropleth map | CalEnviroScreen burden by tract | earthquakes by country | the trap the library already names: large areas dominate. Russia looks seismic because Russia is big |
| the dot density map | 311 request points | earthquake epicentres | CORRECTED against the built files. This row claimed the 311 dots are placed randomly and the epicentres are real. Measurement refuted it — 311 points are 92% unique, geocoded to real addresses. The true pair is better: every USGS event states its own position error (median 0.15 km); the 311 file has no error column at all. One states its uncertainty, one is silent, and silence is not the same as being right |
| proportional symbol map | fire perimeters by size | earthquakes sized by magnitude | area encodes magnitude in both, and magnitude is logarithmic — a circle twice as wide is not twice the quake |
| small multiples | AQI, one panel per monitor | Arctic sea ice, one panel per year | one set is places, one is years. The eye averages panels, and one panel disagreeing is the finding |
| is this real, or noise? | a monthly 311 change somebody claimed | a warming signal against year-to-year swing | the same question at two timescales, and the honest answer differs |
How to run a pair
Not two dives. One dive, two datasets, five minutes.
Same shape, twice. Question 1 is the shape's trap, asked once. Question 2 is: which of these two does the shape serve better, and why?
That second question is the whole mechanism lesson and it does not exist when a shape is taught against one dataset. A student who can say "a dot density map is honest for earthquakes and dishonest for 311 calls, and the reason is where the dots come from" has learned something no single graph could have taught them.
Sources
Civic lane: data.lacity.org · data.lacounty.gov · CalEnviroScreen · Census/ACS for commute and income · LA County canopy and land-surface temperature · CAL FIRE · AQMD · their own collection
Science lane: USGS earthquake catalogue (public CSV, filterable by radius and date) · NOAA GML Mauna Loa CO2 · NSIDC sea ice · NASA/JPL · Gaia or Hipparcos for the HR diagram · NOAA station records
Open
Fifteen shapes × two datasets is thirty CSVs.Term one is BUILT — ten files, five shapes, from USGS, NOAA GML and data.lacity.org, rebuildable withpython3 fetch.py. Provenance and the verified per-pair claims:SOURCES.md.- The HR diagram needs a cut-down file. Full Gaia is millions of rows; a few thousand nearby stars is the teaching version and somebody has to make it.
- Class-collected sets (sleep, lunch) do not exist until a class collects them, which makes them the cheapest and the most scheduling-dependent entries here.