The Sun constantly flares up like a campfire when you toss in more kindling. Big flares are obvious at once, but weak ones easily get lost in the overall glow. The neural network was taught to pick out even barely noticeable bursts from the stream of data that satellites send every second.
From 2018 to August 2025, the algorithm found 111,580 flares. The previous catalog counted only 14,612 over the same period. The difference isn't that the Sun has become more active: it's that new eyes see finer details. The neural network is especially good at finding weak events that used to drown in noise.
When the flares were sorted by strength, they formed a smooth staircase: the weaker the flare, the more of them there are. The old catalog's staircase ended earlier — weak events were missing. The new sample extends it by a full order of magnitude. This means the physics of flares is the same for both giants and dwarfs.
Flares don't follow a strict schedule, but they're not completely chaotic either. Analysis showed that their flow resembles a queue at a coffee shop: sometimes the window serves quickly, sometimes slowly, but on average the process is predictable. Previously, people thought that after a strong flare, the next one waited longer. It turned out to be an illusion: during periods of high activity, weak flares simply got lost, making it seem like the Sun was taking a pause.
The neural network works with data that arrives every second. Over seven and a half years, that's more than 240 million images of the Sun — no team of astronomers could review that many frames.