For four years, NASA’s planet-hunting telescope photographed the light of 2.2 million stars. The data sat in public archives, waiting. Now, a machine-learning system built by astronomers at the University of Warwick has sifted through all of it and pulled out 118 confirmed planets that no one had ever identified, some of them so close to their host stars that they complete a full orbit before a single day passes on Earth.
The findings, accepted for publication in Monthly Notices of the Royal Astronomical Society as of June 2026, also flagged more than 2,000 additional high-probability planet candidates. Nearly 1,000 of those had never appeared in any prior catalog. The total known exoplanet count stands at roughly 5,800 confirmed worlds, so 118 new validations in a single study represents a meaningful jump, and the candidate list could eventually push that number far higher.
What RAVEN found buried in the TESS archive
The Transiting Exoplanet Survey Satellite, or TESS, scans the sky in overlapping strips called sectors, recording brightness measurements for millions of stars. When a planet crosses in front of its star from our vantage point, the star dims by a tiny, repeatable amount. That dip is the signal astronomers hunt for. But with millions of light curves to examine, human reviewers and traditional software can only cover so much ground.
RAVEN, short for RAnking and Validation of ExoplaNets, was designed to close that gap. The pipeline processed full-frame image light curves from TESS sectors 1 through 55, covering roughly the mission’s first four years. It searched for periodic brightness dips with orbital periods between half a day and 16 days, a window that captures everything from scorching ultra-short-period worlds to planets with orbits lasting just over two weeks.
The star sample itself was carefully curated. A separate study cross-matched TESS targets with positional data from the European Space Agency’s Gaia spacecraft, applying quality filters on parallax measurements and surface gravity to build a clean catalog of 2.2 million main-sequence stars. That foundation reduced the risk that stellar variability or light from background objects would mimic a planetary signal, and it gave the pipeline reliable estimates of each star’s size and brightness, which directly determine the inferred size and temperature of any orbiting planet.
Of the 118 newly validated planets, 31 were detected for the first time by RAVEN itself rather than flagged by earlier surveys. The rest had been noted as unconfirmed candidates in previous work but had never cleared the statistical bar for validation until RAVEN’s analysis.
Worlds that orbit in hours, not months
Among the most striking discoveries are ultra-short-period planets, worlds locked in orbits so tight they circle their star in less than 24 hours. To put that in perspective: Mercury, the closest planet to our Sun, takes 88 days to complete one orbit. These newly found worlds are so near their stars that surface temperatures likely reach thousands of degrees, and tidal forces may stretch and deform the planets themselves.
That extreme environment makes them scientifically valuable. Planets this close to their stars are bombarded by radiation intense enough to strip away atmospheres over time, turning them into natural laboratories for studying atmospheric loss, surface composition, and the physics of close-in orbits. According to the University of Warwick’s announcement, many of the new planets orbit small, cool M-dwarf stars. Around those dimmer stars, even a tight orbit can place a planet closer to the temperate zone where liquid water could theoretically exist, though confirming habitability would require far more data than TESS alone can provide.
How the validation works, and where it stops
RAVEN does not confirm planets the way a radial-velocity spectrograph does, by measuring the gravitational wobble a planet induces in its star. Instead, it uses statistical validation: a framework that weighs the probability of a genuine planet against every plausible false-positive scenario, including eclipsing binary stars, instrumental glitches, and blended light from background sources.
The pipeline’s technical architecture, detailed in a companion methods paper, combines Bayesian validation logic with machine-learning classifiers trained to distinguish real transits from imposters. Candidates that exceed strict posterior-probability thresholds and pass radius constraints earn the label “validated planet.” Those that fall short remain candidates, statistically promising but not yet confirmed.
That distinction matters. “Validated” means the odds overwhelmingly favor a real planet once all modeled false-positive scenarios are accounted for. It does not mean anyone has measured the planet’s mass or taken a spectrum of its atmosphere. Detailed characterization of individual worlds will require follow-up with ground-based spectrographs or space observatories like the James Webb Space Telescope. Telescope time is fiercely competitive, so only a fraction of the 118 planets are likely to receive that attention in the near term.
RAVEN is not working alone
The Warwick team’s pipeline is part of a broader shift toward automated planet discovery. NASA’s own ExoMiner model, developed at the Ames Research Center, has already validated hundreds of exoplanets using deep learning applied to both Kepler and TESS data. ExoMiner’s code is open-source, and NASA has described its goal as scaling discovery across the full archive rather than relying on manual vetting alone.
The two pipelines are complementary rather than competing. Each uses different training data, different classification architectures, and different validation criteria, which means their results can cross-check and refine one another. When independent systems converge on the same signal, confidence in that signal rises. When they disagree, it highlights cases that deserve closer scrutiny. The era of a single team painstakingly vetting candidates one by one is giving way to an ecosystem of automated tools that can process entire survey archives in weeks.
What still needs answering
Several open questions surround the RAVEN results. NASA has not issued a statement specifically addressing the Warwick-led findings, and it remains unclear whether the agency will fold RAVEN’s candidate list into the official TESS Objects of Interest catalog or treat it as an independent community product. That distinction affects how quickly these targets filter into broader observation planning.
The candidate pool of more than 2,000 signals also carries inherent uncertainty. Some will eventually be promoted to validated planets. Others will turn out to be eclipsing binaries, instrumental artifacts, or blended background sources. Many will linger in an unresolved state for years, especially those orbiting faint stars that are difficult to observe from the ground.
How many of these worlds sit in habitable zones is perhaps the most tantalizing unknown. RAVEN’s search window of 0.5 to 16 days biases the sample toward close-in orbits, which around Sun-like stars means temperatures far too extreme for liquid water. Around cooler M-dwarfs, though, a 16-day orbit can land squarely in the temperate zone. The preprint does not break out habitability estimates, and any such analysis would depend on refined stellar data from future Gaia releases and assumptions about atmospheric composition that simply do not exist yet.
Why the next big exoplanet discovery may come from data we already have
The strongest evidence behind these findings rests on two primary-source papers that together form a reproducible chain: the results paper provides the full planet list, detection metrics, and validation statistics, while the methods paper lays out RAVEN’s classifier training, its treatment of stellar variability, and the Bayesian priors used to weigh planetary versus non-planetary explanations. Other teams can audit, extend, or adapt the work to additional TESS sectors or entirely different surveys.
TESS is still collecting data. Its extended mission continues to add new sectors, and each one brings fresh light curves that pipelines like RAVEN and ExoMiner can process. The 118 planets announced here came from the first four years of observations. The next four could yield a comparable haul, or a larger one, as the algorithms improve and the baseline of observations grows longer, making it easier to detect planets with slightly wider orbits.
Some of these worlds will eventually be studied in detail. A handful may turn out to be rocky planets in temperate orbits around quiet stars, the kind of targets that drive the search for biosignatures. Most will be too hot, too large, or too faint for current instruments to characterize fully. But collectively, they fill in the map of planetary diversity in our galaxy, one transit dip at a time, and they make the case that the next major discovery in exoplanet science may not come from a bigger telescope but from a smarter algorithm pointed at data we already have.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
