A minimal, randomly initialized model that produces embeddings without any training signal behind its weights. It behaves unpredictably and its outputs carry no semantic meaning — essentially a placeholder or scaffolding tool. Useful for testing pipeline infrastructure without invoking a real model.