An amateur astronomer used Claude Code to surface a TESS planet candidate: classic transit search found it, AI wrote the code and tried to debunk it, and it is not yet on ExoFOP

Why it matters

【Key points】

On Oct 7, amateur astronomer Pavel Rabtsevich posted on X that he had found a previously unreported planet candidate in public NASA TESS data: the star TIC 4206066 dims every 3.18 days, consistent with a body about 1.4 Earth radii, roughly 116 light-years away. The post has about 780,000 views, and "people are finding planets with Claude" spread widely.

By his own later account, an LLM did not find the signal. He used Claude Code to write a standard BLS transit search (not ML) over 120,000+ stars in late-2025 TESS data; models came in afterwards, spending two weeks of checks trying to prove the signal wrong. He says he mainly used Opus 5.5 and Fable 5.1, with GPT-6 Astra and Sol helping on review. His theory for why pipelines missed it: about 6-9 sigma in 2020 data, near NASA's 7.1-sigma threshold.

【Verifiable】

Original thread: https://x.com/p_rabtsevich/status/2107803696058822742

His method clarification: https://x.com/p_rabtsevich/status/2108290948174344198

ExoFOP page for TIC 4206066: https://exofop.ipac.caltech.edu/tess/target.php?id=4206066 (as of Oct 9, 0 TOIs and 0 CTOIs)

Heat: about 810 related posts on X in 24 hours, peaking at 118 an hour.

【Our read】

1. "AI found a planet" is inaccurate. The bigger signal is that coding agents let amateurs build near-professional pipelines on public archives and then use models as adversarial reviewers. What dropped is the barrier to entry.

2. A candidate is not a planet. Watch whether it enters ExoFOP as a TOI or CTOI and whether independent false-positive vetting and follow-up observations happen. We have not seen an official source for claims that NASA approved follow-up; token and naming-rights hype around the ID is unrelated to the science.

3. If this spreads, astronomy's bottleneck shifts from finding signals to vetting candidates.