Last released Oct 5, 2026
Run the same coding task N times per agent and measure how consistent and how expensive it is.