Static training weights refer to the fixed state of model parameters established during pre-training, which define an LLM's unprompted recall prior to any real-time search grounding.