← All use cases

Use case P9 · Human

Hobbyist uses Claude Code to search seven years of NASA TESS data and reports a planet candidate

Pavel Rabtsevich (@p_rabtsevich) ·

The short version

What they did

Pavel Rabtsevich had Claude Code write and run a search across public TESS photometry, with Codex used to cross-check, and wrote up two transit-like signals around one nearby star as a preprint (doi.org/10.5281/zenodo.22967456). It remains a candidate: TESS re-observes the star in November 2026.

Result (as claimed)

“I think I found a planet nobody knew existed. It’s been hiding in NASA telescope data for seven years.” Gizmodo quotes him: “If the star dims on time in November, this gets a lot more serious.”

Prompt

I’m a hobbyist learning to search public NASA TESS brightness measurements for possible planet transits. My starting target is [known planet and host star], followed by [short star list or region]. My project folder is [folder].

1. Explain the method and common false signals in plain words.
2. Write scripts to download the known star’s data and recover its documented dips and period as a first check.
3. If that works, search the short list, compare findings with known planet and candidate records, and test alternative explanations such as eclipsing stars or instrument errors.

Give me plots, scripts and a log of data sources, settings and checks so someone else can repeat the work. Report uncertain signals as candidates. Don’t claim a discovery or say false positives are ruled out without evidence.

Do it yourself

These steps are my suggested way in, not Pavel’s setup.

Level: Advanced. You write and run code against real telescope data, and the hard part is ruling out false signals.

What you need

  • A coding agent such as Claude Code, and Python on your computer.
  • Free access to TESS light curves through NASA’s MAST archive, for example with the lightkurve Python package.
  • A second model or a second person to check the results.

Start simple

  1. Ask the AI to explain transits, light curves and the common false positives in plain terms.
  2. Have it download the light curve for one star that already has a known planet.
  3. Ask it to find that planet’s dips and period. If it can’t, your method isn’t ready.
  4. Only then point it at a short list of stars and log every test.

Take it further

  • Add checks for eclipsing binaries, odd-even depth differences and nearby bright stars.
  • Have a second model review the code and the logs, as Pavel did with Codex.
  • Join a citizen-science project such as Planet Hunters TESS to see how professionals vet candidates.

Check your work

  • Compare every signal against known planet and candidate lists before you get excited.
  • Ask the AI how each signal could be fake, then test those ideas.
  • Call it a candidate, not a planet. Confirmation needs more observations and expert review.

Source checked. The link, author, date and quote were confirmed against the source. The result itself has not been reproduced. Prompt adapted for this library, inspired by the case. Business size: Individual. Results are quoted as the source claimed them and never improved. Spot an error or a broken link? Email charlie@cbarmorecpa.com.

More in Reading, writing and learning

New cases go out every Saturday in Inside AI.