The planets have spent four and a half billion years remaking themselves. Their interiors have melted, their surfaces have been resurfaced, and whatever they were originally built from has been thoroughly overwritten. The small bodies have not. They are the unprocessed material the planets formed from, and they still carry the record.

That record is readable in three ways. Where they are tells us how the giant planets moved, because the gaps and clusters in their orbits are the fingerprints of resonances that Jupiter and Neptune swept through the disk. What they are made of tells us the temperature at which they condensed, and how far material was transported. And how they spin tells us whether they are solid rocks or loose piles of rubble — which decides what happens if we ever need to move one.

The catch is that we know where far better than we know what. Orbits are essentially complete: 99.71% of the catalog has a brightness and an orbit. But only 8.97% have a measured diameter, 2.22% a rotation period, and 0.14% any measurement of what their surface is made of.

All 1,557,369 objects, placed where they were at a single instant, as a rotatable point cloud that runs in the browser with no plugin and no server. Isolate any population, tilt the belt on edge to see its thickness, and watch Jupiter's Trojan camps separate out 60 degrees ahead of and behind the planet.

To be clear about what is ours and what is not: the measurements are not ours. Every orbit, brightness, diameter and spectral class comes from NASA/JPL's Small-Body Database, which in turn aggregates work by thousands of astronomers and the Minor Planet Center. What we built is the database around them — the schema, the taxonomy, the ingest that resolves the source's quirks, and the tools that mine the result. 1,557,369 objects, snapshot 2026-07-29, rebuilt from scratch each release.

The guiding rule is that the semantic model must not replicate the distribution format: every flag-dependent field is resolved into explicit columns at ingest, and nothing unrecognised is silently dropped.

Two examples of why that matters. The near-Earth flag means different things for asteroids and comets, and a blank value means "not applicable" in one case and "never assessed" in the other — so the catalog records 37,506 objects as hazard not determined rather than quietly counting them as safe. And the two spectral taxonomies in common use have opposite grammars: in the Tholen system multiple letters mean an ambiguous classification, while in the Bus system they name a single subclass. Reading one with the other's rule invents data.

The pipeline has three stages and no hidden state. A snapshot is downloaded verbatim and hashed; the build parses it into a semantic model; validation re-checks the result and writes a report that leads with whatever it could not understand. Same snapshot plus same code gives the same database, every time.

Whether an AI collaborator can do real scientific software engineering — not autocomplete, but design decisions, data modelling, numerical work, and the judgment to say "this source field does not mean what it looks like".

The interesting evidence is not the volume of code but the mistakes that got caught. A diagram that rendered at two points and was unreadable. A canvas silently drawing into a 300×150 box in the corner of the screen. An orbit check that failed until it was traced to four-significant-figure rounding in the source data rather than a bug in the code. An image ID that turned out to be a photograph of Martian sand dunes instead of the asteroid it claimed to be. Each was found by looking at the output rather than trusting it.

The verification story matters more than the line count. Positions in the 3-D viewer were checked object by object against JPL Horizons. Jupiter's Trojans emerging in two camps at exactly ±60° of the planet's longitude was never programmed — it falls out of the orbital mechanics, which makes it a test that breaks loudly if the pipeline is wrong. Claims on this site are meant to be checkable, and where something is uncertain it is labelled uncertain.

Rocks in Space is a personal research and education project by John Rachlin, a faculty member at the Khoury College of Computer Sciences, Northeastern University, observing from Lowell Observatory in Flagstaff, Arizona (AAVSO observer code RJOJ). It grew out of The Variable Zoo Project when it became clear that the same instruments and the same pipeline could be pointed at targets that move.

The aims are threefold: build an open, well-modelled catalog of the solar system's small bodies; contribute real measurements where the professional surveys have left gaps; and make the whole thing legible to students and to anyone curious enough to click through. Everything here is free to read and free to reuse with attribution.

Get in touch

Corrections, questions, and collaboration are all welcome — particularly from anyone with a telescope who wants to help fill in the gold bars in the chart above. Write to [email protected].

Where the data comes from

This site talks about "the database", so it is worth being precise about what that means. Rocks in Space holds no original measurements. The database is an integration of public sources:

• NASA/JPL Small-Body Database — orbital elements, absolute magnitudes, diameters, albedos, rotation periods and spectral classifications for every object. This is the backbone, and it is itself a compilation of work by observatories worldwide.
• Minor Planet Center — the designations, numbering and discovery record that give each object its identity.
• JPL Horizons — planetary positions, used for the reference frame in the 3-D viewer.
• NASA PDS Asteroid Lightcurve Data Base and ALCDEF archive — the published rotation periods, amplitudes and the time-series photometry plotted on the observing page, each session credited to the observer who made it.

What is ours is the schema and the integration: a semantic model that resolves the sources' ambiguities instead of copying them, a two-axis taxonomy with hierarchy-aware counts, and the visualisation and data-mining tools built on top. When a page says a figure was derived from our database, it means we mined those public measurements — not that we made them.

Credit where it belongs

Orbital elements and physical parameters come from the NASA/JPL Small-Body Database; planetary positions from JPL Horizons; designations and discovery data from the Minor Planet Center. The physical measurements this catalog aggregates were made by thousands of astronomers over two centuries. This project organises their work; it did not do it.