The data was already there. NASA’s TESS telescope had been watching the sky since 2018, collecting the faint signatures of planets as they crossed in front of their stars. By the end of its fourth year, it had built light curves for over 2.2 million stars. Astronomers had been through much of it. They had found thousands of planet candidates. They believed they had a reasonably complete picture.
They didn’t.
In a study published in the Monthly Notices of the Royal Astronomical Society, a team at the University of Warwick applied a new AI pipeline called RAVEN to the full TESS archive, all 2.2 million stars, covering the telescope’s first four years of operations. The result: 118 newly validated exoplanets, including 31 that had never been detected before, and more than 2,000 high-quality planet candidates, nearly half of them entirely new to the catalog.
RAVEN doesn’t just look harder at the same signals. The pipeline was trained on realistic simulations to learn how to separate genuine planet transits from the noise: binary star systems that mimic the brightness dip of a crossing planet, instrument artifacts, background contamination. Where previous pipelines were designed to run on smaller, targeted subsets, RAVEN was built for industrial-scale processing. It swept through the full archive in one pass and found what focused, piecemeal searches had missed.
What the New Worlds Reveal
Among the finds are two populations that carry scientific weight beyond their count. Ultra-short-period planets are worlds that complete a full orbit in less than 24 hours, closer to their stars than Mercury is to our Sun. These are extreme environments, probably tidally locked and heated to temperatures that vaporize metals. They form a small and poorly understood population: something about the dynamics of early solar system evolution pushes planets to these scorching-close orbits, and every new confirmed example tightens the theoretical constraints on that process.
The second finding is the first direct measurement of just how rare “Neptunian desert” planets are. The Neptunian desert is a gap in the orbital distribution where medium-sized planets, roughly the mass of Neptune, are almost absent around Sun-like stars. The leading explanation is photoevaporation: radiation from the host star strips away a planet’s atmosphere over millions of years, leaving either a bare rocky core or nothing at all. RAVEN’s systematic sweep gave the Warwick team the first unbiased sample large enough to put a number on this scarcity. Neptunian desert planets appear around just 0.08% of Sun-like stars. The desert is real, its edges are measurable, and the number is finally in the literature.
What this discovery points toward is a shift in methodology. TESS has been surveyed in pieces, by different teams, targeting different populations, with different pipelines. Each produced its own catalog, its own detection threshold, its own blind spots. RAVEN ran through all of it with one consistent approach and surfaced 2,000 candidates that previous efforts had left behind. The sheer volume of the search, more than two million stars observed, puts the shortfall in perspective: even a survey this thorough captures only a sliver of the galaxy, and we were still missing planets inside it.
What Comes Next
The team is now working to follow up on the most promising candidates with larger telescopes capable of atmospheric characterization and precise radial velocity measurements. Some of the 2,000 flagged signals will fail under closer scrutiny. Others will be confirmed as real worlds with orbital parameters, spectral signatures, and eventually their own place in the exoplanet catalogs.
RAVEN can also be reapplied as TESS continues releasing new observing sectors. The telescope is still watching. Every new data release is another archive to process, and the pipeline is already built for it. The 31 worlds confirmed for the first time in this study are a beginning, not a ceiling.
Somewhere in the numbers, there are planets circling their stars right now that no one has named yet.
