~/comparisons
Same challenge. Different models.
Every artifact runs right here, with its source code open and the run metadata attached. No ranking and no scoreboard: a record of what each model delivered.
A challenge is a brief. A run is one model answering it inside one harness.
featured challenges
Briefs flagged as featured in the catalog.
3D Rubik's cube simulator in a single HTML file
A 3x3x3 cube in WebGL with Three.js, in a single file with inline CSS and JavaScript. The brief asks for animated layer rotation that never corrupts cube state, the full standard notation (U/D/L/R/F/B with ' and 2), scramble, undo and reset, a move counter, a timer, and both keyboard and button controls. Layer rotation is where the runs diverge most: reparenting cubies onto a pivot and handing them back without accumulating floating-point drift is not trivial.
3D chess game in a single HTML file
An 8x8 board and all 32 pieces rendered in WebGL with Three.js, in a single file with inline CSS and JavaScript. The brief asks for the full rules engine — legal moves, castling, en passant, promotion, check, checkmate and stalemate — plus an orbital camera and a panel showing turn, move history and captures. Three.js comes from the CDN through an inline import map, so the artifact runs inside the viewer's opaque-origin iframe without weakening the sandbox.
matrix
Rows: challenges. Columns: model · harness pairs. An empty cell means that pair did not run that challenge.
scroll horizontally to see more columns
Empty cell: that model · harness pair did not run that challenge.
latest runs
The most recently published runs.