The Transiting Exoplanet Survey Satellite, or TESS, has produced thousands of planet candidates. Turning those candidates into statistically validated planets remains slow, expert-intensive work. Each target requires evidence from multiple instruments, catalogs, models, and quality checks. The challenge is evaluating that evidence at scale without losing scientific judgment or the trail behind each conclusion.
A 41-author team from Caltech/IPAC, the NASA Exoplanet Science Institute, NASA Ames, NASA Goddard, MIT, the University of Tokyo, the Instituto de Astrofísica de Canarias, and other institutions took on that bottleneck. Michael Collier of Holaxis is the corresponding author; Brian Derfer of Holaxis is a co-author.
The resulting study reports 80 statistically validated planets, including 64 newly validated planets. By our count, it is one of the largest studies of TESS data to date.
From hundreds of candidates to 64 new worlds
The study began with hundreds of TESS Objects of Interest. Its largely automated pipeline brought observations, catalogs, transit modeling, statistical validation, quality checks, and expert review into one traceable process. The accepted catalog contains 80 statistically validated planets. Sixty-four had not been validated previously.
Accepted study at a glance
Accepted manuscript
80
Statistically validated
64
Newly validated
41
Contributing authors
Beyond the headline count, the study shows how strongly follow-up imaging shapes which small planets can be validated.
High-resolution imaging changes the result
Statistical validation asks whether a transit signal is much more likely to be a planet than an astrophysical false positive. Nearby or bound stars can mimic or dilute that signal. High-resolution imaging provides contrast curves that constrain which nearby companions could still be hiding in the data.
The team tested the effect directly by running the validation both with and without the available contrast curves. Among the 68 validated planets with contrast curves, 49 would fail validation without the imaging constraints. The effect is strongest for small planets: every validated planet below 1.7 Earth radii would fail without imaging, compared with one-third above 4 Earth radii.
Scaling a painstaking workflow
The validation process combined transit recovery and detrending, signal searches, global transit modeling, automated vetting, repeated TRICERATOPS false-positive-probability runs, companion-catalog checks, and final time-domain modeling. The workflow preserved a conservative manual vetting step before the final catalog.
The system is largely automated, not fully automated. Models and software handle the work that becomes punishing at this scale. Scientists set the thresholds, inspect the evidence, resolve edge cases, and take responsibility for the result.
Each result remains traceable through the computations, observations, assumptions, and checks that produced it. The study also provides full machine-readable tables.
How people and AI agents contributed
The paper describes the scientific methods and findings. The accepted manuscript acknowledges that agents from Anthropic and OpenAI assisted across a roughly four-month window with analysis code, computational workflows, figures, tables, references, and manuscript work. The authors reviewed and verified every calculation, interpretation, figure, table, reference, and passage of manuscript text.
Agent speed is valuable only inside a system that preserves evidence, makes assumptions inspectable, and keeps expert judgment in control.
Why this is a Holaxis proof point
This work began in astronomy, but the underlying problem is broader. Serious knowledge work accumulates data, decisions, exceptions, methods, and evidence over time. Teams need durable structures to verify results, improve the process, and begin the next problem from a stronger starting point.
The TESS study demonstrates a different pattern:
- Experts define what a trustworthy result requires.
- People and agents explore, implement, test, and revise the workflow together.
- Validated methods become reusable infrastructure.
- The system preserves enough evidence for others to inspect and reproduce the work.
- Each resolved problem strengthens the starting point for the next one.
Together, people and agents carried a difficult scientific program through to a rigorous, inspectable result. Just as importantly, the methods and infrastructure remain available for the work that comes next.
Read the accepted study
The manuscript has been accepted for publication in The Astronomical Journal, and the accepted preprint is available on arXiv. We will add the journal DOI and final publication details when they become available.