TESS photographs most of the sky every 27 days. The full-frame images contain millions of stars, each a potential exoplanet host. Finding transits — the tiny, periodic dips in brightness when a planet crosses its star — in this data requires inspecting millions of light curves. Human vetting doesn't scale.
Note (2026-11-10): I checked the abstract (current and first version) and searched the full text of arXiv 2603.22597 (Lafarga et al.). The scope is narrower than the essay's: RAVEN works on a magnitude-limited sample of over 2.2 million main-sequence stars characterized by Gaia, in TESS-SPOC full-frame images from sectors 1 to 55, for periods of 0.5 to 16 days. It detects candidates with a box least squares search, classifies them with machine learning models trained on realistic simulations, and performs statistical validation. It newly validates 118 planets, 31 of them newly detected here, and presents over 2,000 high-probability candidates not validated. The essay's list of checks (an odd-even transit comparison and secondary eclipse checks) is not what the paper describes; I did not find an odd-even comparison, and the centroid information enters through a positional probability. The essay's claims about the cost of the 101st planet and about the bottleneck moving from detection to validation are its own.
The authors (arXiv:2603.22597) build RAVEN, an automated pipeline that processes TESS full-frame images end to end: extract light curves, search for periodic signals, classify candidates, and vet for false positives. Over 100 newly validated planets and over 2,000 vetted candidates.
The validation is the hard part. A transit signal can be mimicked by eclipsing binary stars, instrumental systematics, or stellar variability. RAVEN combines machine learning classification with centroid analysis (is the dip centered on the target star?), odd-even transit comparison (are alternating transits identical?), and secondary eclipse checks (does the companion emit light?).
The through-claim: the bottleneck in exoplanet discovery shifted from detection to validation. TESS detects thousands of signals; the question is which ones are real planets. Automated validation at scale produces diminishing returns per candidate but increasing returns per survey — the 101st planet costs the same to validate as the 1st, but each new planet refines the population statistics that tell us how common Earth-like worlds are.