In 2018, volunteers classified 55,000 candidate variable stars; 860,000 classifications helped flag one that became about 100 times fainter

Someplace in the crowded middle of the Milky Way, a star has been fading. Not exploding, not vanishing in a single dramatic flash, but slowly, steadily dimming over years until it was roughly a hundred times fainter than when anyone started watching it. Not one of the major observatories flagged it first, but a swarm of ordinary people, peering at brightness graphs on their laptops.

That's the quiet reward of a citizen-science project called Variable Star Zoo, launched in 2018 by Chile's Millennium Institute of Astrophysics (MAS). According to the project's own accounts, more than 5,000 volunteers from the US, Europe, India, Brazil, Mexico and Canada entered over 860,000 star classifications in under two years, searching for oddities buried in a much larger database. One of those oddities was that fading star. Its nature remains unknown to astronomers.

When thousands of strangers do what computers can't


This worked because people are, for certain tasks, simply better than software. The input list came from the VVV survey, a long-running scan of the galaxy's inner bulge. The project's own background material, summarized by the MAS team in a 2022 paper titled “Infrared variability of young solar analogues in the Lagoon Nebula,” by Ordenes-Huanca et al. and published in the Monthly Notices of the Royal Astronomical Society, puts that list at roughly 55,000 candidate variable stars, each one a brightness graph waiting to be sorted into a category, or flagged as something nobody had a category for yet. Machine-learning algorithms are decent at recognizing patterns they've already seen, but poor at recognizing the pattern they've never encountered, because they don't know how to look for it.

A similar but otherwise independent project, Citizen ASAS-SN, ran into exactly that wall. According to Christy et al., in their 2022 study titled “Citizen ASAS-SN Data Release. I. Variable Star Classification Using Citizen Science,” published in Publications of the Astronomical Society of the Pacific, automated pipelines routinely fail to pick up rare or unusual variable stars precisely because they're rare, while human volunteers are especially good at spotting the outlier that doesn't fit any known bucket. That's essentially what happened here: among the tens of thousands of stars the volunteers were asked to label, their human eyes picked out roughly 8,000 old, rapidly pulsing stars that the project's own computer algorithms had only caught about half of.

The star that kept fading, and nobody has an answer yet

Buried in that pile was one star that didn't behave like any of the templates volunteers were shown: not an eclipsing binary, not a Cepheid, not a Mira variable. It did fall into the catch-all "unusual" category volunteers could use for exactly this kind of oddity, but nobody, then or since, has been able to say what it actually is. It simply kept getting dimmer, observation after observation, until it settled at roughly one percent of its original brightness. The MAS team has been upfront that they don't know what's causing it; it could be a simple eclipse by an unseen companion, a shroud of dust drifting between us and the star, or something rarer. Without more follow-up observation, it's an open question rather than a headline-generating mystery.

The catch: crowds aren't perfect either

It's worth resisting the urge to romanticize the crowd too much; citizen classifications aren't uniformly reliable. A large-scale analysis of a similar project, the SuperWASP Variable Stars effort, found that volunteer-labeled light curves were about 89 percent accurate for eclipsing binary stars, but only around 9 percent accurate for rotationally modulated variables, according to a 2021 paper titled “SuperWASP variable stars: classifying light curves using citizen science,” also published in the Monthly Notices of the Royal Astronomical Society.

In other words, people are very good at spotting the clearly weird thing, and considerably less good at fine-grained sorting. The strength of projects like Variable Star Zoo isn't that volunteers replace professional astronomers, but that they act as a first filter, wide enough to catch what algorithms might miss, before scientists take over with actual telescopes. It is a modest but genuinely useful role, and it means that somewhere out there, a person who has never touched a research grant or a graduate degree saw a star behaving strangely, years before anyone with a PhD did.