Artifacts
What each model built on each task, shown on the held-out cases it was graded on.
Sentinel‑2 gap filling
Each model's best and median solver, re-run on hidden test tiles (the ones the reward is computed on) with whole dates withheld: what it painted in, how far its NDVI is from the truth, and how unsure it said it was. Public validation tiles from Portugal show the same, but agents could tune on them, so they are not graded; click a pixel to follow it through the season.