4 610 universes — that's how many computer models astronomers needed to show that the familiar 'Gaussian mask' no longer fits next-generation data. This mask is applied when you assume all uncertainties follow the same law — the normal distribution (a smooth bell-shaped curve). But in space, things are different. Dark matter (the invisible stuff that holds galaxies together), first noticed by Vera Rubin, clumps into huge clots. That breaks the simple picture. Future sky surveys, like the Euclid telescope, will measure the large-scale structure of the universe from tiny distortions in the shapes of distant galaxies. From that data you can learn the matter fraction Ωm (how much stuff is in the universe), and thus test the nature of dark energy (the mysterious force that makes the universe expand faster and faster) and how gravity governs this accelerated expansion, discovered by Edwin Hubble and Georges Lemaître.

To extract numbers from those distortions, scientists use two approaches. The classical method (ELI) believes in the Gaussian mask: it thinks any errors can be described by a simple 'bell curve'. For that it uses numerical simulations (computer universes) and builds fast approximate calculations. The alternative method (LFI) takes the mask off. It uses Machine Learning, specifically Convolutional Neural Networks (programs that can find patterns in digital images — sets of pixels). On many simulations, the neural network learns what the data actually looks like, and then, during Bayesian inference (a way to update probabilities as new information comes in), it honestly estimates the needed parameters.

The researchers set up a fair duel across 4 610 universes. The old method, ELI, behaved like a person in a mask trying to guess an object by touch: if the data was simple, it managed, but as soon as a slight 'irregularity' appeared, it made mistakes. For example, when using a neural network to compress data, ELI's predictions for the key parameter S8 (a measure of matter clumpiness) were 2.2 times less accurate than LFI's. The new method, LFI, by contrast, always gave correct and precise results.

The main lesson: the Gaussian mask has become a dangerous illusion in the era of precise measurements. The work proved that methods without rigid assumptions are not a toy but a necessity. But it also showed that the old method can work if the data behaves 'by the rules'. The practical takeaway: any future analysis of cosmic data must undergo a strict honesty check — calibration. Otherwise, method artifacts can be mistaken for new laws of nature. By taking the mask off those 4 610 universes, scientists gave us a chance to see the real large-scale structure without distortion.