The branch walking pipeline systematically substitutes high-confidence alternative tokens from logprobs to prompt the model with new prefix fragments, discovering latent associations and secondary brand connections that exceed a specified threshold.