The depths of space are not silent — they are full of encrypted messages. Every star, every quasar sends its flickering light code: rhythmic like the Morse of a Cepheid, or chaotic like a cracked cipher of a supernova. Astronomers intercept these signals by the millions each night — but how to read them without a key? Telescopes like ZTF produce endless streams of light curves, and here machine learning steps onto the stage — not just a listener, but a seasoned cryptanalyst.

Imagine a cryptanalyst handed millions of ciphertexts without a key. He listens to patterns, compares fragments, and gradually the structure of the language emerges before him. That's exactly how Astra-CLR works. Its architecture is a transformer with dual vision: from each light curve, a 'global' frame and several 'local' snippets are created, sometimes noisy, distorted, with gaps. The network learns to match these views so that the hidden essence of the object remains unchanged — like a cryptographer recognizing a message even if it's transmitted with interference.

The result is akin to the insight of a codebreaker who pieces the common code back together bit by bit. After pre-training on 2.1 million light curves, the model mastered an 'alphabet' of 12 variability classes: from eclipsing binaries to cataclysmic objects with white dwarfs. A simple logistic regression on the extracted vectors gave 70% accuracy — as if the decoder confidently tells a diplomatic cipher from a trade one. And a tiny bit of labeling (just 2% of the data) raised accuracy to 77%, collapsing the latent space into tight clusters — now the analyst discerns the individual styles of the senders.

This approach is not just a classifier — it's a universal 'embedder', a ready foundation for any future surveys. LSST and Roman will flood us with data where manual labeling becomes a luxury. Astra-CLR shows that you can train a neural network to see the essence of things without hints, just by showing it the world — and then, with a slight twist of the knobs, turn it into a precise instrument for scientific discovery.

The next step is to teach the network to combine time and color in dynamic codes and transfer learned knowledge across different surveys. This is the path to a universal decoder of cosmic signals, able to spot anomalies in real time. Perhaps someday such a cryptanalyst will notice the first hint of unknown physics — and the quiet whisper of spacetime curvature will turn into readable text. Interestingly, the clustering of representations sometimes reveals subgroups of objects that could point to new astrophysical phenomena — potential 'strange attractors' of stellar dynamics, awaiting theoretical interpretation.