Princeton's T16 Pipeline Mines NASA's TESS Archive for 10,091 New Exoplanet Candidates, More Than Doubling the Mission's Tally
A Princeton-led reanalysis of 83.7 million TESS light curves used random forest classifiers to surface 11,554 planet candidates, including one confirmed hot Jupiter, by reaching stars 16 times fainter than prior searches.
Overview
A team of astronomers at Princeton University has released the largest single-pass exoplanet candidate catalog ever drawn from NASA’s Transiting Exoplanet Survey Satellite (TESS), surfacing 11,554 transit signals — 10,091 of them previously undetected — from a fresh reduction of the mission’s first year of full-frame images. The work, posted to arXiv on April 20, 2026 and titled The T16 Planet Hunt, more than doubles TESS’s previous candidate count and demonstrates that machine-learning classifiers can extract planets from the faint end of the survey that earlier pipelines could not reach.
What We Know
The T16 pipeline, named for its target magnitude limit, processed 83,717,159 TESS Cycle 1 light curves down to T=16 — stars that are 16 times fainter than the brightness limit used in most prior systematic searches, according to Universe Today’s coverage of the project. The reanalysis pulled out 11,554 candidate signals with orbital periods between 0.5 and 27 days, and the authors explicitly state that the haul “more than double[s] the number of known TESS exoplanet candidates,” as detailed in the arXiv preprint.
Classification of the candidates was handled by random forest classifiers, with separate models trained for bright (T<14.5) and faint (T≥14.5) stars to optimize performance across the magnitude range, the paper reports. The pipeline also flagged 411 single-transit events — long-period candidates that show only one dip during the observing window — a category that historically requires targeted follow-up to confirm.
To prove the catalog contains genuine planets and not just statistical noise, the team validated one signal end-to-end. Using radial-velocity measurements from the Planet Finder Spectrograph on the 6.5-meter Magellan telescopes at Las Campanas Observatory in Chile, the arXiv paper confirms a hot Jupiter orbiting TIC 183374187 — a metal-poor, thick-disk star — with a mass of 0.56 ± 0.19 Jupiter masses and a 5.059-day orbital period.
The lead author is Joshua T. Roth of Princeton’s Department of Astrophysical Sciences, with co-authors Joel D. Hartman and Gáspár Á. Bakos and collaborators from the Carnegie Institution and other institutions, per the arXiv author list. Bakos’s group has long built automated photometric surveys, and the T16 effort is in keeping with that lineage of large-scale, pipeline-driven planet hunting.
What We Don’t Know
The overwhelming majority of the new candidates remain unconfirmed. Only one — the hot Jupiter around TIC 183374187 — has been validated with follow-up spectroscopy in this paper. Confirming the rest will require radial-velocity work, ground-based photometry, or independent statistical validation, and the false-positive rate at faint magnitudes is historically higher than in the bright-star samples that earlier TESS pipelines focused on.
The authors describe a mixed candidate population. While hot Jupiters are a prominent type in the sample, the arXiv paper reports that the transit search also surfaced warm Jupiters with orbital periods between 10 and 100 days, Neptunes, and potential sub-Neptunes and super-Earths. How many small, rocky, or temperate worlds ultimately survive deeper vetting from the 11,554 signals will only become clear after follow-up spectroscopy and statistical validation.
The paper also leaves open how much of the catalog will translate into formally confirmed planets in the long run. TESS had already produced roughly 8,000 candidates and about 760 confirmed planets prior to this work, the preprint notes; whether the T16 sample yields a similar confirmation ratio, or a lower one because of its fainter targets, is a question for the next several years of follow-up programs.
Analysis
The T16 release lands in a year already shaped by industrial-scale planet hunting. The Machine Herald has previously reported on JWST’s atmospheric measurements of the forbidden exoplanet TOI-5205b and on NASA’s unveiling of the fully assembled Nancy Grace Roman Space Telescope, which is expected to register tens of thousands of new microlensing planet detections after its fall 2026 launch. The T16 catalog shows that even before Roman flies, classical archive mining — combined with modern classifiers — can reshape the known exoplanet inventory using data already on disk.
The choice of random forest classifiers rather than deep neural networks is notable. Random forests are interpretable, fast to train, and resistant to the kind of overfitting that can plague transit searches at low signal-to-noise. By splitting the bright and faint populations into separate models, the Princeton team effectively built two specialized detectors, allowing the faint-star pipeline to chase candidates that brightness-cut surveys would have discarded as noise.
Whether the broader field follows T16’s playbook — reanalyzing existing TESS, Kepler, and K2 data with magnitude-aware classifiers — will determine how many more “hidden” planets remain in archived photometry. With Cycles 2 through 5 of TESS still to be processed at this depth, the 11,554-candidate figure may turn out to be a floor rather than a ceiling.