Public Frontier: an open laboratory for finding new worlds with AI and public data

I.Introduction

On 25 September 2026, Pavel Rabtsevich, an independent researcher in Tarragona, Spain, posted a preprint about TIC 4206066, a nearby late-K dwarf star.[1][1] Rabtsevich, P. (2026). Two Transit-like Signals in TESS Photometry of the Nearby K Dwarf TIC 4206066. Preprint, Zenodo. doi.org In data from NASA's Transiting Exoplanet Survey Satellite (TESS), he found a dip in the star's brightness that repeats every 3.18 days. It is about 500 parts per million deep and shows up in 23 transits across three observing sectors. If a planet causes it, and if that planet orbits this star, the planet is about 1.4 times the radius of Earth. The note does not claim more than the data support. It names the main alternative, a planet around an unseen companion star, and states that no statistical validation is claimed. Claude Code, an AI coding assistant, helped with the code, calculations, figures and drafting under his direction. The full data and code are public.[2][2] Rabtsevich, P. (2026). Supporting data and code for "Two Transit-like Signals in TESS Photometry of the Nearby K Dwarf TIC 4206066". Zenodo. doi.org

Now the claim faces a clean test. The TESS mission granted two-minute observations of the star for its Sector 110, which runs from 31 October to 26 November 2026.[3][3] TESS Science Office (2026). Director's Discretionary Targets, program 100: two-minute cadence for TIC 4206066. tess.mit.edu Before those data exist, Rabtsevich published the exact transit times he expects and the rule he will use to judge the result.[4][4] Rabtsevich, P. (2026). A Pre-registered TESS Sector 110 Test of Two Transit-like Signals in TIC 4206066. Zenodo. doi.org When those data reach the public archive, anyone can check whether the dips come back on schedule.

Amateurs and volunteers have found real things in public astronomy data for years, as our review of that history shows. What has changed is that a careful person can now write and run a professional-grade analysis with an AI assistant, on data that is free to download. What has not changed is that most signals turn out to be something else, such as eclipsing binary stars, light from a neighbouring star or quirks of the instrument. The work that matters is the checking.

II.How this lab works

Public Frontier is built around one loop. You can join it at any step.

1. Learn on a known planet. The tutorial downloads real TESS data and finds a confirmed planet from scratch, step by step, with a notebook you can run.
2. Search. Point the same method at stars nobody has looked at closely. An AI assistant can write the code, but you should be able to explain every line.
3. Check. Compare odd and even transits, look for a secondary eclipse, look for contaminating neighbours, and search the catalogues to see if the object is already known.
4. Share. Publish a short note with your figures, measurements, code and data, and say how you used AI. Mark it honestly as a candidate, a recovery of a known object, or a false positive.
5. Discuss. Others rerun your notebook, question the steps and suggest what would confirm or rule out the signal. When something is published or confirmed, the author updates the note.

Notes and comments are stored as public records in each author's own ATProto repository, the open protocol behind Bluesky. This site only reads and displays them. You keep your work if this site disappears, and anyone can build another view of the same records.

III.Recent discoveries

IV.Start here

If you have never looked at a light curve, start with Find your first planet. It recovers the giant planet HD 2685 b from two sectors of TESS data. The notebook runs in a few minutes in Google Colab with nothing to install, downloads about 85 MB, and ends with numbers you measured yourself. Every figure and number in the tutorial came from running that code on NASA data.

V.Research programmes

VI.Principles

• Show your work. Link the code and data that produced every figure. A claim nobody else can rerun is not worth much.
• Candidates are candidates. Call a signal a planet only after independent follow-up confirms it. "Candidate" is an honourable word.
• Negative results count. Showing that a signal is an eclipsing binary saves everyone else the time.
• Say how you used AI. Which tools, for what, and how you checked their output. Using AI is welcome here. Hiding it is not.
• Look before you announce. Check ExoFOP, the NASA Exoplanet Archive and SIMBAD first, using this checklist. Recovering a known object is good practice; announcing it as new is embarrassing.
• Your records are yours. Notes and comments live in your own repository. You can edit or delete them at any time.

Acknowledgments

This site includes data collected with the TESS mission, obtained from the MAST data archive at the Space Telescope Science Institute (STScI). Funding for US Institutions for the TESS mission is provided by the NASA Explorer Program. STScI is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5-26555.[5][5] Mikulski Archive for Space Telescopes (2026). Mission acknowledgements: TESS. Space Telescope Science Institute. archive.stsci.edu The tutorial relies on Lightkurve and Astropy. Notes are stored as standard.site documents with Leaflet content blocks. The drawings are ink studies made with an image model from public-domain references and historical telescope drawings.