Tech »  Hallucination by Design: How Embedding Models Misunderstand Language