fchollet/arc-agi

ARC-AGI-1 is a benchmark dataset plus a browser tool for manually solving its grid tasks. A user loads a task, reviews training demonstrations and test inputs, edits an output grid with resize, copy, selection, and floodfill tools, then submits it for exact-answer validation and moves through remaining test inputs. The task corpus has separate training and evaluation sets. The excerpts establish grid rendering and state synchronization, but omit some handlers and the HTML wiring, so only documented or directly evidenced relationships are shown.