An interpreter in a room where three languages are spoken simultaneously first hears only a general noise. But if they start asking clarifying questions, listening to intonations, and repeatedly confirming who answered with which phrase, individual voices gradually emerge from the cacophony. This is roughly how the new algorithm, Solar-CDC, works: it has for the first time separated the slow solar wind into two populations that are indistinguishable by speed but differ in origin.

The solar wind is a stream of charged particles that the Sun continuously ejects into space. The fast wind (about 700 km/s) is confidently linked to coronal holes — regions where magnetic field lines are open and particles freely escape. The slow wind (about 400 km/s), however, could originate both in the streamer belt, a zone of closed magnetic loops, and at the boundaries of the same coronal holes. The problem is that both sources produce plasma at nearly the same speed: the difference between them is only 4 km/s, like the difference between a pedestrian's pace and a light jog. But the composition of heavy ions, which 'freezes' near the Sun and remains unchanged thereafter, differs between them. Yet how can one separate the streams if the boundary in the multidimensional composition space is not visible to the eye?

Solar-CDC acts differently. It takes seven plasma characteristics (speed, density, charge-state ratios of oxygen and carbon, the iron-to-oxygen ratio, and the mean charge of oxygen) and first roughly sorts the observations into three clusters. Then a neural network learns to rearrange the data so that points from the same cluster move closer together, while those from different clusters repel each other. Every few steps, the clusters are reassessed, and the cycle repeats until the picture stabilizes. This resembles an interpreter who does not just listen but actively asks: 'You pronounced this word with such an accent, so you must be from that group?'

The result was clear-cut. On 30,602 observations from Solar Orbiter, the algorithm found three populations. Two of them are slow wind with different compositions: one has an oxygen ion ratio O7+/O6+ of 0.080, the other 0.400—that is, 2.5 times higher. The third population is the fast wind. Moreover, an intermediate group with a value of 0.160 corresponds exactly to the boundaries of coronal holes. Even if the key variable, that very oxygen ratio, is removed from the data, the algorithm still recovers the separation, albeit less accurately.

Now that the two 'voices' of the slow wind are distinguishable, one can trace how each of them affects space weather near Earth. After all, disturbances from different sources carry different consequences for satellites and communications. And the method itself is not tied to the Sun: it can separate hidden populations in any data where the signal is distributed across many variables — from spectra of distant stars to cell analyses. An interpreter who knows how to ask clarifying questions can be seated at any negotiation table.