Two hundred forty-two thousand three hundred twenty-seven galaxies — like grains of sand on a beach. But astronomers turned this chaos into a map: the UMAP algorithm (a machine learning method) took from each galaxy eleven parameters — seven colors from optical to mid-infrared (that's photometry), magnitude, radius, and photometric redshift — and reduced them to two coordinates. On a two-dimensional sheet, similar objects lay next to each other, forming archipelagos and straits. Just 68 spectroscopy confirmed Little Red Dots — compact galaxy at high redshift — became reliable beacons, outlining the islands where these mysterious sources hide.

The resulting map showed: Red Dots clustered into two dense clumps. The main archipelago — 56 objects with a median redshift z≃5,1 — at completeness 0,82 showed purity 0,78. This is comparable to the best photometry selections, but without a single explicit color threshold. No color boundaries. Their spectra speak for themselves. The second, smaller island of 11 objects sits at z≃3,4. There, in the archival data, four new active galactic nucleus with broad Hα line were found — a signature of fast gas around a black hole.

Inside these islands, the algorithm found 107 new Red Dot candidates missed by previous catalogs. They are compact and lie on a characteristic V-shaped continuum in the rest frame, but slightly bluer than the strictest color cuts. Their nature is not yet clear: possibly black hole hidden by gas and cosmic dust, or dusty bursts of star formation. Notably, not a single brown dwarf — a traditional nuisance in color selections — entered the LRD regions. These failed stars gathered on a separate island with blue photometry F115W−F200W, completely cut off from the Red Dot archipelago.

The method is not tied to photometry. The same framework can be applied to spectroscopy and high-resolution images to search for subtle spectral features or morphological subclasses. An interactive web application has already been created — a living atlas of distant Universe populations, where anyone can explore the map, images, and spectra. In the future, such maps will help process data from the Euclid and Roman telescopes, accelerating the discovery of rare object classes. Like a real island map, it will be supplemented with new archipelagos, and each new atlas will open uncharted lands in the ocean of data — without a single colored flashlight.