The night sky is dotted with stars, and many of them have planets. But how do you spot a tiny dark disk against the glare of a distant sun? The Kepler telescope spent nine years watching hundreds of thousands of stars, waiting for a planet to pass in front of its star and briefly dim it. That 'wink' is the only way to find worlds like ours.

The trouble is that at the edge of sensitivity, the telescope starts seeing things that aren't there. A speck of dust on a lens, a jitter of the spacecraft, random electronic noise — all of these can mimic a tiny eclipse. Previously, such false signals were weeded out by hand, but for the faintest and rarest events, there was no reliable filter. Scientists built a new processing pipeline: first, the images are cleaned of local defects, like a restorer removing scratches from an old photograph, then each individual brightness dip is checked on its own, not in bulk. Instead of a simple yes/no, they use a Bayesian statistical test that compares how much the data resemble a real planet versus random noise.

Then the new pipeline was applied to stars with already known candidates. It successfully returned all previously confirmed planets to their places. But many unconfirmed candidates, especially those with long orbits and weak signals, it flagged as false alarms. Among them are several 'Earth twins' in the habitable zone: KOI 8063.01, 8107.01, and 8242.01. They can no longer be considered planets.

This changes an important number: how often Earth-sized planets occur in the habitable zone. If some of the earlier finds were mirages, the estimate becomes more modest. But don't be sad: next-generation telescopes are already looking at the sky, and there will be plenty of real worlds to find.