A developer reported finding a planet not listed in existing catalogs by feeding public astronomical data into Claude Code. The post appeared on Reddit and was discussed on Hacker News, where it received 49 points and 14 comments.
The approach relied on Claude writing and debugging Python scripts that processed light curve data and orbital parameters. No custom hardware or large datasets were required beyond files already available from public archives.
How the Analysis Worked
Claude generated code to load FITS files, apply period-finding algorithms, and flag transit signals that did not match known objects. The model iterated on the script after the user supplied error messages and output plots.
The process stayed within standard Python libraries such as NumPy, pandas, and lightkurve. Each refinement cycle took minutes rather than hours of manual coding.
Community Reaction on Hacker News
Commenters noted the 49-point score and asked for the exact dataset and verification steps. Several users requested the full conversation log with Claude to assess whether the detection holds under peer review.
Others pointed out that similar finds have occurred with simpler scripts, but the speed of iteration with an LLM lowered the barrier for non-specialists.
How to Replicate the Workflow
Users can start at claude.ai, upload a CSV or FITS excerpt, and prompt the model to write a transit search script. Provide the error traceback after the first run to trigger fixes.
Public data sources include the NASA Exoplanet Archive at https://exoplanetarchive.ipac.caltech.edu and Kepler light curves hosted on MAST. The original discussion thread is available at https://www.reddit.com/r/ClaudeAI/s/mbe5IY2LF9.
Pros and Cons
- Pros: Rapid script iteration, handles boilerplate astronomy code, works with modest local hardware.
- Cons: Model can hallucinate orbital parameters; final confirmation still requires manual vetting and follow-up observations.
Alternatives and Comparison
| Tool | Iteration Speed | Astronomy Libraries | Verification Needed |
|---|---|---|---|
| Claude Code | High | Good | High |
| GPT-4o | High | Good | High |
| Manual coding | Low | Full control | Medium |
Claude and GPT-4o both reduce initial coding time compared with writing everything from scratch, but neither replaces statistical validation.
Who Should Try This
Amateur astronomers with basic Python experience and access to public survey data will see the largest gain. Professional researchers already using established pipelines will gain less unless they need quick exploratory scripts.
Bottom Line
Claude Code lowered the time to produce a candidate signal from public data, yet independent confirmation remains essential before any claim is accepted.
Top comments (0)