Hunt for an exoplanet in public telescope data and win observing time

A Reddit user browsing NASA TESS data with Claude Code noticed a star with a repeating dip. Over about two weeks they directed the agent to download and parse the data, write a transit search, fit the transits, check neighbouring stars and other false-positive signals, plot results and re-run anything that failed, while they set the questions and pass/fail rules. Codex and fresh read-only Claude sessions reviewed the work independently, and the agent searched dozens of catalogues and hundreds of thousands of alerts to confirm it was unreported. A 3.18-day signal held up across 2018, 2020 and 2025 data and in hold-out tests; one overstated claim caught in audit was withdrawn. A TESS observing proposal to test it was approved.