A scatter plot gives you two dimensions. Add colour and you get three. Add size and you get four. At that point most visualisation tools tap out.
But a 3D sphere in a navigable space has many more properties than that:
That's nine independent data dimensions encoded in a single object. And you're not looking at it from a fixed viewpoint — you can fly through it, orbit it, approach it from any angle, stand inside it.
This is the idea behind the DataViz tool we built on Saturday night.
The Tool
Drop in a CSV. Any CSV. The tool reads the column headers, computes min/max/mean for each numeric column, and presents you with six mapping selectors:
- X axis — any column
- Y axis — any column
- Z axis — any column
- Size — any column
- Colour — any column
- Opacity — any column
Change any selector and the point cloud reorganises itself in real time. Four transition modes: Instant, Fast (400ms), Slow (1.8s), and Drift (3.5s with smooth easing). In Drift mode you can watch the relationships between dimensions physically rearrange themselves as the mapping shifts. Patterns that were invisible in one configuration suddenly crystallise in another.
Five colour palettes: Thermal (black→red→orange→white), Stellar (red M-stars through blue O-stars), Spectrum, Ocean, and Mono.
Hover any point to see all its column values. The column statistics panel shows the range of every dimension at a glance.
The Galaxy Test
Before we could trust this tool with real data, we needed to validate it with something we already know the shape of.
We generated a synthetic Milky Way: 5,000 stars, procedurally placed according to known galactic structure.
- 4 spiral arms — logarithmic spirals with scatter, young hot blue-white stars concentrated in the arms
- Thin disk — exponential radial distribution, older orange-yellow stars
- Bulge — prolate spheroid at the centre, K/M giant dominated, warm orange
- Halo — sparse spherical distribution, old metal-poor stars, semi-transparent
Stellar colours are physically motivated — temperature drives colour through the OBAFGKM sequence. Luminosity drives size through a compressed main-sequence scaling. Halo stars are transparent to show they're background objects.
The test: if the encoding is honest, the structure should be self-evident.
Edge-on view: a thin lens with a central thickening. Exactly right. The classic galaxy silhouette.
Top-down view: spiral arm structure visible as overdensities. The bulge bright at centre.
Inside the disk: stars in every direction, denser toward the core, the disk rising around you like a landscape. This is the view our ancestors had of the Milky Way before they knew what it was — a band of light across the sky that is, in fact, the disk seen edge-on from inside.
No 2D projection can give you that view. You have to be inside it.
The pipeline is validated.
Why This Matters for GearSense
GearSense measures instantaneous gear ratio by dividing wheel frequency by crank frequency — a single derived value that places each moment of a ride into one of a discrete set of possible gears.
But a ride isn't just gear ratios. It's cadence, speed, power, heart rate, elevation, time. Each measurement is a dimension. The interesting questions live in the relationships between dimensions — does my cadence cluster differently when I'm on flat vs climbing? Does heart rate track power linearly or does it decouple at high speeds? Where does the motor assistance kick in and what does that do to the gear ratio signal?
A scatter plot answers one question at a time. The DataViz tool answers all of them simultaneously, from every angle.
The first synthetic GearSense dataset — 200 data points, 8 dimensions — is already loading in the tool. Swap cadence to X, gear ratio to Y, speed to Z, power to colour, heart rate to opacity, and what you see is not random. There is shape. There is structure. The data has something to say.
We just needed a way to listen to all of it at once.
What Comes Next
- Export to VRML — so any mapping can be loaded into the full VRML viewer for fly-through navigation with the viewpoint system
- Real GearSense ride data — actual commute data from the EnergyOne
- Gaia DR3 stellar catalogue — 50,000 real stars, see if the structure self-assembles
- Axis labels in 3D space
- Point selection and comparison
The tool and the galaxy benchmark are open source under TIC-OHSL. They live at git.theidea.club alongside the VRML viewer.
Built with Claude Sonnet 4.6 and Three.js r128 on a Saturday night, after Tombo Chan flew again for the first time in thirty years.
Loading…