TESS

A home search for planets in NASA's TESS data, and the one candidate it found so far.

NASA’s TESS satellite watches most of the sky, star by star, in roughly month-long stretches. When a planet crosses in front of its star, the star dims by a tiny, regular amount. The data are public. So in October 2026 I started searching them on my own hardware at home, on Cortex, the same machine that runs HKOS.

The point is the same as everything else here: can a person with a home computer, open data and careful methods do real work, and be honest about where it stops?

What it found

One candidate. A dip of about 196 parts per million every 5.450 days on TOI-913 (TIC 407126408), a K dwarf 65 parsecs away that already has one known planet. If the signal is on that star, the planet is about 1.1 times Earth’s radius, orbiting inside the known planet’s orbit.

It is a candidate, not a discovery. It also wasn’t first: the open-source eumemic/exoplanets search detected it two days before I did. What I’ve added is an independent detection plus vetting, and we compare notes in public on their GitHub.

What it survives

  • Nine of ten automated vetting tests (the tenth, on transit shape, is inconclusive).
  • A season holdout: split the data into four observing seasons, and the transit shows up in three of them on its own. The fourth falls just short.
  • A false-alarm test on 200 shuffled copies of the light curve: under a 1.5% chance it’s noise.
  • An upside-down copy of the light curve, which shows how strong a peak the instrument’s quirks make on their own. The real signal beats the strongest one by a wide margin.
  • Recovery in a second, independent set of TESS photometry.

What is weak

  • A faint star 27 arcseconds away (TIC 407126406) could be an eclipsing binary faking the dip. Our own centroid test only weakly disfavours it. That star is the main reason this isn’t a “likely planet” yet. The target sits at declination −81°, so a ground check needs a southern telescope.
  • My pipeline has blind spots. It found 23 of 26 known planets at the right period, but it missed 7 of 17 of eumemic’s signals in a blind re-check.
  • No professional astronomer has checked it. I built it with a lot of help from an AI assistant (Claude), directed by me, and I checked the key numbers by hand.

The real test

TESS looks at this star again in Sector 114 (February–March 2027). Before that data exists, I published the times the transits should land (pre-registration, doi:10.5281/zenodo.23290795). Within 30 days of the data, I’ll publish what it shows, whatever it shows.

Read more