The optical sky background is a cosmic palimpsest, where light from billions of unresolved galaxies is layered, each layer corresponding to a different epoch. Until now we could only distinguish the brightest "letters"—individual stars and nearby galaxies. But the bulk of the text remained a blurry smudge, concealing the history of the expansion of the universe, the birth of the first luminaries, and the nature of dark matter. Unlike the cosmic microwave background, which preserves an imprint of hot plasma, the optical background is a stratified chronicle, where each layer corresponds to an era of stellar nucleosynthesis. A new method, akin to spectral analysis of ancient inks, makes it possible to separate these layers and recover the redshift distribution of galaxies — a key quantity linking observed light to distance and time.

Researchers trained a neural network, conditional normalizing flows, on realistic simulations of multiband photometry from the Euclid space mission and the ground-based LSST, where each pixel stores the combined brightness of hundreds of galaxies. Imagine trying to reconstruct the age structure of a forest from a single blurry photo of a green canopy: different tree species (galaxies at different z) have slightly different hues, and only by flipping through many filters can you guess where young saplings are and where ancient oaks stand. Nancy Grace Roman, the visionary of space telescopes, laid the groundwork for this "color" archaeology, while Edwin Hubble gave us a ruler for measuring cosmic distances — Hubble's law, linking redshift to distance.

The method's precision is striking: the mean redshift and its dispersion are determined with an error of less than one percent even when a fifth of the data is missing. The network learned to self-calibrate: if noise is added to the image, the prediction of the noise's standard deviation immediately signals an unreliable result. Like an experienced conservator cleaning a palimpsest knows when a layer is damaged and doesn't jump to conclusions. Tomographic analysis, splitting into four redshift intervals, confirmed that we can confidently slice the glow into epochs, except for edge effects where the voices are very faint.

This opens the door to a new kind of observational cosmology. Previously unresolvable galaxies become tracers of large-scale structure, allowing tests of hypotheses about the nature of dark matter and the rate of star formation in the young universe. Vera Rubin, whose rotation curves of spiral galaxies pointed to hidden mass, would likely have appreciated a tool that can pick up its signature in background fluctuations. Looking ahead — integration with radio observations, hydrodynamic numerical simulations, and cross-correlation with catalogs of resolved objects promises a breakthrough in understanding the early universe.

Thus, "blind" mapping of the background turns into full-fledged 3D cartography of cosmic history. Several long-standing puzzles are tackled at once: from the "missing" ultraviolet radiation to the role of interstellar dust that masks primeval light. Next steps are real data from Euclid and the Roman Space Telescope, where each pixel will be not just noise but a page of chronicle awaiting reading. And then the palimpsest metaphor will gain physical precision: beneath the later overleafs, the era of the first stars will emerge, and perhaps even traces of the dark ages themselves.