A research-backed workbench for turning the web's best AI coding prompts into inspectable SVG instruments, browser games, 3D studies, and simulations you can actually use.
Every record keeps the original idea, the evidence attached to it, the runtime result, and the revision that changed it.
Browse public examples from model launches, developer posts, repositories, and benchmark harnesses.
Choose a render family, attach sources, and get a safe iframe with controls, telemetry, and a real reset path.
Revise the build, replay its hash chain, and export a Markdown report with provenance and an honest score.
GitHub and DEV results are normalized into the same shape. When providers fail, the fallback is labeled instead of pretending to be live.