News · 2026-10-09

TESS approves observations of a hobbyist’s Claude-assisted planet candidate

The Transiting Exoplanet Survey Satellite, known as TESS, approved new observations of TIC 4206066, a star around which hobbyist Pavel Rabtsevich reports finding a transit-like signal with substantial help from Claude Code. The official target list calls the object a planet candidate and schedules two-minute observations under discretionary observing program DDT 100. That validates an observation request, while the underlying analysis and the existence of a planet remain unconfirmed.

Key facts

• Rabtsevich reports a signal repeating every 3.18 days, with an estimated size of 1.4 Earth radii if a planet causes it.
• TESS lists active DDT 100 with Rabtsevich as principal investigator and TIC 4206066 as the target.
• The author says the new observation window runs October 31 through November 26, 2026; those dates are not in the official target table.
• The independent primary anchor is TESS’s Director’s Discretionary Targets page.

The strongest part of this story is a test that has not yet happened. A person working with coding agents has made a prediction about a repeating astronomical signal, and an established observatory has approved additional measurements. That makes the next data more informative than another round of social-media enthusiasm.

In his first-person ClaudeAI post, Rabtsevich says he noticed the signal in late-2025 TESS data, then found matching dips in observations from 2020 and 2018. He describes a brightness decrease of roughly 0.05 percent lasting about two hours. His own description is “planet candidate, not a confirmed planet.” The analysis was not independently rerun during the dossier research.

TESS looks for small, repeated changes in a star’s brightness that may indicate a planet passing in front of it. Imagine watching a distant lamp while a tiny object occasionally crosses the beam. The repeating shadow helps identify a possible orbit, but it does not establish which object produced the shadow or whether the lamp you intended to watch was actually the source.

That distinction matters with astronomical images containing multiple sources. A nearby eclipsing binary can create a misleading signal. Instrumental effects and incorrect estimates of the host star can also change the interpretation. TESS’s follow-up program describes imaging, spectroscopy, and radial-velocity measurements used to investigate false positives and characterize candidates.

Rabtsevich reports that Claude Code, using Opus 5.5 and Fable 5.1, performed much of the execution: downloading and parsing public data, writing searches, fitting possible transits, checking nearby stars, producing plots, and repeating analyses. He says he selected the hypotheses, tests, and pass-or-fail criteria. Codex and fresh read-only agent sessions reportedly reviewed the work. This is an account of assisted analysis, not evidence that a model independently conceived and validated a discovery.

One reported check held out a year of observations while fitting the other two years, then tested whether the predicted dips appeared in the held-out year. He says all three rotations of that test succeeded. The logic resembles cross-validation: an explanation gains credibility when it predicts data that did not determine the fit. Its value still depends on correct implementation and on whether the held-out data were truly kept separate from earlier choices.

The author says a review caught an overstrong statistical claim, which he removed and recalculated while retaining the main signal. That is a useful workflow detail, but the corrected result remains his report. Multiple agent reviewers can find errors without becoming a substitute for independent astronomical reproduction.

He also reports a weaker signal near 11.13 days, conditionally associated with a larger companion. It should not be promoted into a second discovered planet. The main candidate and the weaker signal have different evidential status. A singularity repost intensified the framing into a two-candidate discovery and a claim that NASA would confirm it in November. Approved photometry supports a scheduled test, not that promised outcome.

The official DDT page confirms the target, cadence, investigator, and active status. It also says approved observations have no proprietary period. Rabtsevich says he published transit predictions and decision rules before the upcoming observations, but the linked archive records were inaccessible in the dossier. The preregistration’s contents therefore were not independently checked. The principle is still worth understanding through preregistration.

Comments in the original thread include both encouragement and technically specific skepticism about borderline signal strength and a possible background source. The commenters claiming astronomical expertise are anonymous and their credentials were not verified. No named astronomer’s independent endorsement was established. The appropriate anticipation is clear: future dips at the predicted times would strengthen an interesting candidate, while confirmation requires more than an approved observation slot.

Primary source, verified: read the paper →

Ground Truth. (2026, October 9). TESS approves observations of a hobbyist’s Claude-assisted planet candidate. Ground Truth. https://groundtruth.day/news/tess-approves-observations-of-claude-assisted-planet-candidate.html

@misc{groundtruth:tess-approves-observations-of-claude-assisted-planet-candidate,
  title  = {TESS approves observations of a hobbyist’s Claude-assisted planet candidate},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {oct},
  url    = {https://groundtruth.day/news/tess-approves-observations-of-claude-assisted-planet-candidate.html}
}

Topics: ai-science · astronomy · claude-code · citizen-science · verification