“The Speculative-Speculative Decoding (SSD) algorithm predicts likely verification outcomes by using the token distributions from the draft model, specifically considering tokens that were plausible but not sampled.”