ML finds about 10,000 new planet candidates

Joshua T. Roth and colleagues used machine learning to comb through archival data from NASA’s Transiting Exoplanet Survey Satellite, or TESS, and assembled a catalog of 11,554 planet candidates, according to a paper published April 28 in *The Astrophysical Journal Supplement Series*. The team said 10,091 of those signals had not been flagged before, after analyzing 83,717,159 light curves from TESS’s first year of full-frame observations. The search focused on stars as faint as magnitude 16 in the TESS band, extending well beyond the brighter targets emphasized by many earlier TESS planet searches. ### How did the researchers find so many candidates in old data? The T16 project built a “uniformly detrended and systematics-corrected” set of TESS Cycle 1 light curves and then ran a semi-automated transit search across the full sample, the paper said. In transit searches, astronomers look for repeated dips in a star’s brightness that can indicate a planet passing in front of it. The authors said machine learning mattered because the scale of the dataset was too large for conventional manual vetting alone. (arxiv.org) Their search used TESS full-frame images rather than only the mission’s standard target lists, letting the team probe many fainter stars that official pipelines did not prioritize. ### What exactly is in the new catalog? The paper said the catalog contains 11,554 planet candidates with orbital periods between 0.5 and 27 days. (arxiv.org) Of those, 10,091 are newly identified candidates, 1,052 correspond to previously known TESS candidates, and 411 are single-transit events for which the team did not assign orbital parameters. The authors wrote that the findings “more than double” the number of known TESS exoplanet candidates. (arxiv.org) For comparison, the NASA Exoplanet Archive listed 7,931 TESS project candidates as of May 15, 2026, and 895 confirmed TESS planets as of May 21, 2026. The archive’s TESS candidate table is built from the Exoplanet Follow-up Observing Program, or ExoFOP, which is updated twice a day. ### Why were faint stars a big part of the search? (arxiv.org) The T16 paper said most existing TESS planet searches have focused on relatively bright targets, even though planet occurrence rates imply many planets should also orbit fainter stars. By pushing down to TESS magnitude 16, the search widened the target pool substantially. Phys.org, citing the study, said the project processed stars “16 times fainter” than those typically targeted by official TESS searches. (arxiv.org) That broader sweep does not confirm planets on its own, but it gives follow-up teams a larger list of targets to test with radial-velocity measurements, additional transit observations and catalog cross-checks. ### Did the team confirm any planet from the new search? (arxiv.org) Magellan/PFS radial-velocity observations confirmed one newly identified signal as a hot Jupiter orbiting TIC 183374187, the paper said. The authors described the host as a metal-poor thick-disk star and said the confirmation was meant to test whether the pipeline could surface real, previously undiscovered planets. That single confirmation does not validate the full catalog, but it provides one worked example of how the candidate list is supposed to be used: machine learning identifies likely signals first, then telescopes and follow-up teams sort planets from false positives. (phys.org) That is an inference from the paper’s workflow and the archive’s role in maintaining candidate tables. ### What happens next for the 10,000 new signals? (arxiv.org) The April 28 paper said the candidate list is a target set for future validation and follow-up efforts, especially around faint stars. The next step is not another algorithmic pass alone, but observations by named facilities and programs that can refine or rule out the signals. The NASA Exoplanet Archive and ExoFOP-TESS are the places where many of those follow-up results are likely to surface as candidates are updated, reclassified or confirmed. (arxiv.org) As of May 21, 2026, the archive was still adding new planets, spectra and system data in its regular updates. (exoplanetarchive.ipac.caltech.edu)