GPU-Resident Speculative Decoding for In-Browser LLM Inference with WebGPU

Every version keeps its own PDF, citation, and review record — endorsements are bound to the version they examined and never carry forward silently.