Where the data comes from

What is inside the atlas right now. Every number below is checked on every change by an automatic test: if a catalogue changes shape underneath us, the test turns red before anyone sees a star in the wrong place.

Stars

HYG Database 4.1, Astronexus — Hipparcos, Gliese and the Yale Bright Star Catalogue combined · CC BY-SA 4.0

The catalogue holds 119,626 stars; 109,400 have a usable distance and 95,987 are in the atlas. The 13,413 missing from that count are not an error: they are out by a declared choice of scale — everything within 150 parsecs gets in, and beyond that only stars brighter than magnitude 9.2, out to 3,000 parsecs. The remaining 10,226 have no usable distance, and without a distance there is nowhere to put them.

For each one we have magnitude, colour index and catalogue number: the colour you see on screen comes from the star’s real temperature, not from a styling decision.

Extrasolar planets

NASA Exoplanet Archive, pscomppars table, NASA/IPAC · public domain, citation requested

2,413 confirmed planets in 1,709 systems within 200 parsecs, with semi-major axis, period, radius and mass taken from the papers of the people who found them. None excluded: the selection is already in the table, which holds confirmed planets only.

The solar system

JPL, Approximate Positions of the Major Planets (Standish) · moon elements from JPL Horizons · public domain

Nine planets in their true positions for the date on screen, and 21 moons. Being partly transcribed by hand they are the part most exposed to a typo, so they are the part with the most checks: Earth must sit at 1.000000 AU, the Jovian year at 4332.6 days, Europa at 3.551181 days, Venus and Triton must turn the wrong way round. A check like that already caught an error in Europa’s period that nobody had spotted by looking.

The surfaces you land on

NASA Trek · NASA GIBS · Terrain Tiles · public works

Going down onto the Moon, Mars and Earth, the imagery is not inside the program: it arrives as tiles while you descend, from the servers of the people who measured it.

• The Moon — the global LRO WAC mosaic at 303 pixels per degree, from the Lunar Reconnaissance Orbiter.
• Mars — the Viking MDIM21 colour mosaic at 232 metres per pixel.
• Earth — VIIRS Corrected Reflectance in true colour, 250 metres, from the Suomi NPP satellite. It is not today’s Earth: the archive publishes a few days behind, so the atlas looks up the most recent date that answers instead of guessing it, and writes that date on Earth’s card. It used to ask for “yesterday”, nothing arrived, and the lens stayed empty with nobody noticing.
• The relief, and for Earth only — Terrain Tiles (SRTM, ETOPO1, GMTED2010), an open dataset hosted on Amazon S3. It stops at zoom level 10, and that limit is measured: over the Mariana Trench the elevation reads −10,434 m up to there and exactly zero beyond it, because the bathymetry ends. Reading that zero would give a flat, believable ocean precisely where you fly low over the water. The Moon and Mars have no elevation source: they stay spheres, and the card says so. Without that line, someone descending onto the Moon after descending onto Earth would see a painted sphere and could not tell whether the Moon is smooth or whether it is we who do not know.

It also means that, as you land, your browser talks to those servers: it is written on the privacy page.

The maps of bodies seen from a distance

Solar System Scope · Creative Commons Attribution 4.0

The maps of the Sun, Mercury, Venus, Earth (day, night and clouds), the Moon, Mars, Jupiter, Saturn with its rings, Uranus and Neptune come from solarsystemscope.com/textures. The licence asks for attribution, which is why this line also appears inside the atlas, next to the images themselves. The textures the source itself marks as “fictional” were not used: they would be invented surfaces on real bodies, which is precisely the thing this project does not do.

Deep sky and galaxies

positions from catalogue, distances from the literature, transcribed by hand

31 deep-sky objects and 35 galaxies, from the satellites of the Milky Way out to the Coma Cluster at 99 megaparsecs, which is the most distant measured object you can reach today. The positions come from the catalogue; the distances and diameters were taken one by one from the literature, because the source catalogue has no distance column.

Sagittarius A*

GRAVITY collaboration

The black hole at the centre of the Galaxy: mass and distance from the GRAVITY measurements, obtained by following for years the orbits of the stars that go round it.

What you see is not a photograph. The famous Event Horizon Telescope image, from 2022, exists and is cited in our data, but it is not shown here: in the atlas Sagittarius A* is drawn. At the distance you look at it from, a photograph and a drawing are indistinguishable — which is exactly why it has to be said which of the two you are looking at.

Downloaded, counted, not yet visible. These catalogues are on disk and the compilation work is under way. The figures are here, and none of them ever appears without its status: a bare number on this page would read as “this is inside the atlas”, and that would be a promise rather than a measurement. Withholding the number, though, would hide the gap behind a silence — and that is exactly what is not done here: what is measured is declared, what is generated is declared, and what is missing is declared too.

Downloaded and counted, none of them visible in the atlas yet

• Gaia Catalogue of Nearby Stars — Gaia Collaboration, Smart et al. 2021, A&A 649, A6, CC BY 4.0. 331,312 stars within 100 parsecs, all admitted. The radial velocity, though, exists for only 82,360 of them, a quarter: without it you know where a star is but not where it is going.
• Million Quasars v8 — Flesch 2023. The name belongs to the catalogue; it is not a count. 1,012,615 objects downloaded and checked column by column, of which 1,012,614 have a measured redshift and none visible in the atlas yet. But the real quasars number 860,100: the other 152,514 are active nuclei, Seyfert-2s, radio sources and X-ray sources. They exist, they have a redshift, and they are not quasars — calling them that would be convenient and false.
• Cosmicflows-4 — Tully et al. 2023, ApJ 944, 94. 55,877 galaxy distances, all admitted.
• 2MASS Redshift Survey — Huchra et al. 2012, ApJS 199, 26. 44,599 downloaded, 43,415 admitted. The 1,184 excluded have a redshift corresponding to less than 500 km/s, and below that threshold redshift is not a distance: what dominates it is the local motion of nearby galaxies.
• Open clusters — Hunt & Reffert 2023, A&A 673, A114. 7,167, all admitted.
• Globular clusters — Harris 1996, 2010 edition. 147, of which 145 have a distance.
• Pulsars — ATNF Pulsar Catalogue, Manchester et al. 2005, CSIRO. 2,536, and here the difference matters more than the total: 83 have a parallax, that is, a geometric distance. The other 2,398 have a distance inferred from how much the signal was slowed by gas along the way, plus a model of how that gas is spread through the Galaxy: 20 to 50 per cent of error. Both kinds will get in, but they say two different things and the card will write which. 55 have no distance at all.
• Abell galaxy clusters — Abell, Corwin & Olowin 1989. 2,712, of which 568 have a redshift. The other 2,144 have only a distance class estimated from the magnitude of the tenth brightest galaxy: that is not a measurement, and they do not get in.
• Small bodies — JPL Small-Body Database, NASA/JPL-Caltech, public domain. 139,583 asteroids with a measured diameter, 8,340 trans-Neptunian objects, 1,089 centaurs, 4,088 comets. 1,440,911 asteroids stay out — and not because the data is missing: their orbits are measured very well. It is a declared choice of ours, because without a diameter a body cannot be drawn at its true size.

The 168,412 galaxies that exist and whose place we do not know

The deep-sky catalogue holds 222,949 galaxies. Cross-matching them, within 30 arcseconds, against the redshift catalogues 2MRS and Cosmicflows-4, 54,537 gain a measured distance and will get in.

The other 168,412 have a name, a direction in the sky and a magnitude. They have no depth. Putting them into the scene would mean inventing their third coordinate, and it would be invisible: nobody looking could tell.

We have not thrown them away and we have not put them in. They are waiting in a file, and they will arrive on their own the day a redshift catalogue reaches them.

What is generated instead. It is here because the most honest part of an atlas is the list of what it has not measured.

The shape of the Milky Way beyond ~3,000 parsecs

No catalogue holds the individual stars at those distances. The disc, the arms, the bar and the halo are built from published density models: the overall shape is the one the research describes, but the individual stars you see out there do not exist one by one.

The deep field beyond ~150 megaparsecs

Statistically plausible nodes and filaments, not observed galaxies. That is why they cannot be selected.

Hypothetical systems

Around stars with no confirmed planets the atlas shows a system generated from the star’s catalogue number — so always the same one for that star, never random — and the card marks it as hypothetical.

The engine, and what it asks of the network

three.js r128 draws the scene, typography by IBM Plex. Neither of them lives in this folder: three.js comes from cdnjs.cloudflare.com and the typefaces from Google Fonts. Those are two external services, and so they see the IP address of anyone opening the page. They will have to come in here; until they do, the privacy page says in full what leaves and who receives it.