Direct discovery · Physics
Claude-assisted pipeline releases a UV sky map with explicit measured and predicted layers
A public ultraviolet sky mosaic combining archival surveys with flagged statistical predictions, provenance weights and uncertainty layers.
Summary
Ménard reports a multi-day Claude Science collaboration that calibrated archival UV surveys and filled missing coverage using other wavelengths. The released project describes gradient-boosted regression, spatial cross-validation and per-pixel provenance; roughly a third of the displayed sky is inferred. Human review caught atmospheric artifacts missed by two agent reviews.
AI role
Agents retrieved surveys, wrote calibration and fusion pipelines, and fitted statistical regressions under astrophysicist guidance.
Narrative role
The inspectable mosaic shows agents completing a substantial astronomy data-integration task with human supervision and explicit uncertainty.
Caveat
Predicted regions are not new observations and cannot establish unseen objects. Effective predicted resolution is 0.5–1 degree; the educational map has no independent scientific validation or measured end-to-end speedup.