How this map is built

Every star and galaxy sits at its real measured position, in one continuous space. There are no separate "levels", you are flying through a single 3D model on a logarithmic scale, so structures nest the way they truly do: the neighborhood inside the galaxy, the galaxy inside the cosmic web.

The stars come from the AT-HYG catalogue (Hipparcos, Tycho and Gaia), nearly 2 million of them with measured distances, colored by their real temperature: hot blue-white giants, our yellow Sun, cool red Antares. They form a bubble a few thousand light years across, the limit of where stellar distances can actually be measured, including the real stars above and below the galactic plane.

The galaxies come from two redshift surveys, 620,033 of them placed by their measured redshift through Hubble's law: the all-sky 2MASS survey for the local sphere, and the deep SDSS survey for the great wedge that reaches out past 500 million light years. The cosmic web, its filaments, walls and voids, is not drawn; it falls out of the data.

Between the galaxy and the deep web sits the Local Group and Local Supercluster: real nearby galaxies at their measured distances, Andromeda, the Magellanic Clouds, Centaurus A, the Whirlpool, and the Virgo Cluster that anchors our supercluster. The bright galaxies here are real, from the 2MASS Redshift Survey, weighted by their measured luminosity so cluster cores glow.

The glowing web at the supercluster scale is a reconstruction. A real survey is a sparse scatter of galaxies, not the continuous filaments you see in simulation images, so the soft web behind the real galaxies is modeled: the bright nodes sit at the true positions of Virgo, Coma, the Great Attractor and the other real clusters, and the filaments and voids between them come from a Voronoi-foam model of the cosmic web, boosted to follow the real 2MASS density. The structure is a plausible reconstruction, not an observation; the real galaxies ride on top of it.

Some things are honest models, not surveys. The planets sit at their real orbital distances, and the asteroid belt, Kuiper belt, heliopause and Oort cloud are drawn at their real distances too, but as representative clouds, we have no catalogue of individual Oort cloud objects. The two Voyager probes sit at their real 2026 positions, the only human-made objects out there, already past the heliopause and climbing out of the ecliptic plane. And the far Milky Way is a schematic spiral, because we sit inside our galaxy and cannot measure most of its stars; the real stars fill the bubble around the Sun, then blend into the model beyond.

The halo is real again. Surrounding the disc, the roughly 150 globular clusters (Harris catalogue) and the dwarf satellite galaxies (Sagittarius, Fornax, Sculptor, Draco, Leo I and dozens more, from the McConnachie catalogue) sit at their real measured distances, filling the space from the disc edge out to the Local Group that would otherwise look empty.

The "AI fill" toggle is different from everything else: it is not real data. A 3D diffusion model I trained on simulated cosmic-web fields dreams up plausible galaxies to fill the sky the surveys never saw, behind the Milky Way and outside the SDSS wedge, so you can imagine what the whole cosmic web might look like. It is statistically web-like but invented; the crimson points are inferred, never observed.

Data: AT-HYG / HYG (astronexus, public domain) · 2MRS (Huchra et al. 2012) · SDSS · globular clusters (Harris 1996, 2010 ed.) · Local Group dwarfs (McConnachie 2012). Rendered with Three.js. Code on GitHub.