Codexstronomy
Grants & community for finding new objects in public sky data.
We're setting up grants and a community for people using AI agents like Codex to find new things in public sky data, with a group chat where people share progress and help each other with the work.
A $50,000 program from macrogrants, seeded by Will DePue. We're accepting project ideas now, and donations to fund more.
To get started, email or text us a short description of what you want to look for.
Want to support with compute, model credits, or a donation? Email us.
An independent macrogrants project, not affiliated with Anthropic, OpenAI, xAI, or Google. Please no cease-and-desist 馃檹. Status and giving.
Why
There's tons of data sitting in public archives: full-frame images, light curves and catalogues, and people are still finding new things in them. A search can miss things outside the parameters or goals it was designed for, and combining archives can turn up something that wasn't clear in either one alone.
Anyone today with a laptop and an agent can help search them again. People already are (see below). Much of the data is free to access! TESS, Gaia, ZTF, DECam, and WISE have public releases, and HSC is available with registration for non-commercial science and education. (We might host mirrors of some of this where the archives allow it.)
We'd like to give compute, Codex, Claude, Grok or Gemini credits and small grants to people who want to do this seriously. Also, if a candidate earns telescope time, we can help you ask for it.
Nulls are really important too! A calibrated non-detection with a completeness curve is a valuable result, as long as we're clear about the search's sensitivity and coverage.
Things we'd like to help people look for
- Things that move. Asteroids, comets, distant solar system bodies, and interstellar visitors. Stacking images along possible tracks can reveal faint moving objects, while linking detections across nights or surveys can establish their motion.
- Things that transit. Planet and eclipsing-binary candidates, including single transits, long periods, faint hosts, and blended neighbours. Agents can help run checks, but candidates still need careful false-positive analysis and, where appropriate, follow-up observations.
- Things that change. Variable stars, flares, outbursts, dimming events, and transients that become easier to characterize when you look across years of observations.
- Things that are nearby. Brown-dwarf and faint-star candidates that reveal themselves through proper motion or parallax across epochs, especially in the infrared.
- Things that become interesting across archives. An infrared source with no apparent optical counterpart, unusual motion, or a transient with detections spread across surveys. These can be leads to investigate; mismatches and missing measurements can also come from the data or the way catalogues were joined.
Public archives
| Archive | What is in it | Where |
|---|---|---|
| TESS | Full-frame images covering most of the sky since 2018, at 30-minute, then 10-minute, and since 2022 200-second cadence; faster light curves for selected targets. | MAST, lightkurve |
| Kepler / K2 | Four years of near-continuous measurements of roughly 150,000 stars, followed by K2 campaigns along the ecliptic. The archive is still producing discoveries years after the telescope's 2018 retirement. | MAST |
| Gaia DR3 | Positions and brightness for about 1.8 billion sources, with parallaxes and proper motions for about 1.46 billion. Colours, spectra, and time series for subsets. | DR3 docs, ESA archive |
| ZTF | Repeated imaging of the northern sky since 2018, typically every few nights, with public releases and an alert stream. | IRSA |
| Pan-STARRS | Five-band imaging of roughly three quarters of the sky, with stacked images and multi-epoch detections from the 2010–2014 survey. | MAST |
| DECam / DES | DES imaging and catalogues cover roughly 5,000 square degrees of southern sky. Public data from other DECam programs are available through NOIRLab's archives. | DES, image archive |
| HSC-SSP | Deep Subaru imaging: PDR3 covers about 670 square degrees in all five broad bands at full depth, and 1,470 square degrees including partial coverage. Registration required; non-commercial scientific or educational use. | HSC |
| WISE / NEOWISE | Repeated all-sky infrared observations from 2010–2011 and 2013–2024, with more than 20 sky passes during NEOWISE. Useful for faint moving objects and variability. | IRSA |
| SDSS | Imaging of more than a third of the sky and millions of optical and infrared spectra, useful for classifying objects and measuring their physical properties. | sdss.org |
Some work that inspires this
These range from citizen science and conventional machine learning to AI-assisted analysis. None were funded by macrogrants. Some are candidates or preliminary reports, as noted.
- An AI-assisted planet candidate around TIC 4206066. Pavel Rabtsevich reports a 3.18-day transit-like signal in public TESS data, consistent with a planet roughly 1.4 times Earth's radius if it is on the target star. He describes using Claude Code and Codex to build and challenge the analysis, including checks against older observations. TESS has approved two-minute observations under DDT Program 100, planned for October–November 2026. The signal is not statistically validated or confirmed as a planet. Author's report, TESS DDT programs.
- 3,006 brown-dwarf candidates from archived infrared images. Backyard Worlds volunteers searched WISE and NEOWISE images for moving sources. An April 2026 paper reports 3,006 motion-confirmed L- and T-dwarf candidates whose spectral types still need follow-up. If confirmed, the sample would more than double the known L and T dwarf population. Paper.
- A microlensing planet in the TESS archive. A team led by Mallory Harris combined a 2023 Gaia alert with archived TESS observations to identify Gaia23bra b, a giant planet nearly 40,000 light-years away. TESS's dense observations revealed features Gaia's sparse sampling missed. Published July 2026. Paper, NASA.
- A giant planet confirmed with help from 24 citizen scientists. After TESS recorded a single transit of TOI-4465 b, radial-velocity observations indicated a roughly 102-day orbit. A worldwide observing campaign, including 24 citizen scientists, captured another transit and helped confirm this roughly six-Jupiter-mass planet. Reported June 2025. Paper.
- 118 newly validated planets from a fresh TESS search. Warwick's RAVEN pipeline searched observations of more than 2.2 million stars from TESS's first four years. The March 2026 study statistically validated 118 planets, 31 newly detected in the search, and reported over 2,000 additional high-probability candidates, about 1,000 of them new. Paper, Warwick.
- 27 planet candidates from changing eclipse times. A team led by Margo Thornton studied 1,590 eclipsing binaries with at least two years of TESS observations. After modeling changes in eclipse timing, they identified 27 systems with possible planets requiring follow-up. Published May 2026. NASA.
- White dwarfs that fade for months or years. A search of ZTF alerts for fading sources identified six white dwarfs with deep obscuration events, interpreted as dust and debris from disrupted planetesimals. One disappeared for more than a year while brightening in the infrared. September 2026 preprint. Preprint.
- Six variable stars from a small urban telescope. Maksym Pyatnytskyy reports six new variable stars found with a 150 mm telescope in Kyiv, then characterized using public TESS data. April 2026 preprint. Preprint.
- About 7,000 TESS targets flagged for further study. NASA Ames' open-source ExoMiner++ model identified about 7,000 TESS targets as exoplanet candidates in an initial run. These signals need vetting and follow-up, which is exactly the kind of work we'd like to support. NASA.
Some things we care about
- Show that your search can recover known objects in the same data before trusting it to find new ones.
- Inject realistic synthetic sources across the parameters you're searching, measure how often you recover them, and calibrate false positives separately.
- Write down thresholds and predictions before new data arrives, and document any later changes or bug fixes.
- Have a fresh agent session challenge your analysis, and ask people who know the data to look too. Separate agents can share the same mistakes.
- Publish code, methods and results so others can reproduce the work. With a candidate, be clear about what has and hasn't been ruled out.
- Publish useful null results too, with enough about sensitivity and coverage for the next person to know what could still have been missed.
Get involved
Email us or text us at (415) 516-3509 with a short description of what you want to look for. If you already know which archive you want to use, what you've tried, or what would help, include that. These are suggestions, not requirements, and there is no application form.
We'd also like to hear from people who want to help with someone else's search. We'll respond as fast as we can.
If you'd like to support the program with compute, model credits or funding, email us. We're interested in support from any provider, and people can use whichever tools work for their research.
Status and giving
Will DePue is seeding this program with $50,000 through macrogrants. We're looking for significant donations, to the program or to specific projects. Email team@macrogrants.com and we'll arrange it directly.
macrogrants is a California nonprofit public benefit corporation, incorporated October 10, 2026. Our application for 501(c)(3) recognition was filed with the IRS on October 13, 2026 and is pending. Under IRS rules, recognition applies retroactively to our incorporation date, so donations made while the application is pending are tax-deductible once it's granted.
This invitation comes from Will DePue, an unpaid volunteer. macrogrants supports science, education, and public-benefit projects. Information about revenue, expenses, administration, and fundraising is available on request by email or at 2108 N St Ste N, Sacramento, CA 95816, and we provide it within seven days.