Claude Science Helps Complete a Full Ultraviolet Sky Map
An astrophysicist used Claude Science agents to harmonize decades of ultraviolet surveys and estimate the third of the sky never observed by a telescope.
Anthropic says Claude Science helped astrophysicist Brice Ménard produce the first detailed ultraviolet map covering the entire sky. The project combines observations from NASA’s GALEX and Swift missions, South Korea’s FIMS/SPEAR instrument, Europe’s TD-1 survey, and additional Gaia, Planck and infrared data.
The challenge was not simply drawing a picture. GALEX mapped roughly two-thirds of the sky between 2003 and 2013 but avoided areas around very bright stars, including much of the Milky Way’s plane. Other surveys filled some gaps, yet the datasets differed in calibration, resolution and observing conditions. Ménard used Claude Science to search for public datasets, standardize them, remove residual artifacts, align them on a common coordinate system and merge them into a single map.
About one-third of the ultraviolet sky has never been directly observed. Claude estimated those regions through an inpainting process trained on the relationship between ultraviolet measurements and observations at visible, infrared and radio wavelengths. The team tested the method by hiding known ultraviolet regions and asking the system to reconstruct them; the predictions came within about 10% of the measured values in the reported validation.
The workflow was not autonomous in the strongest sense. Ménard set the scientific direction, inspected outputs and identified errors. In one example, faint circular patterns remained from individual GALEX exposures because atmospheric glow had not been fully removed. After the issue was pointed out, Claude traced the artifact to its source and recalibrated all 38,000 observations.
The result is primarily an educational and visualization resource, not a new telescope survey. Predicted pixels are labelled and uncertainty estimates are included, which is essential because a model-generated estimate cannot substitute for direct observation. Its significance lies in showing how agentic systems can complete neglected, technically tedious scientific infrastructure projects while leaving the scientist responsible for interpretation and validation.