Pre-registration of a replication of the inference-latency claims of REVBRAIN (arXiv 2610.38558)

Frozen pre-registration (1 Nov 2026) of a private replication of two quantitative claims of arXiv 2610.38558 (Pankka, Lehtinen, Roine): that the median inference latency of the autoregressive WaveNet baseline is about 52x that of REVBRAIN at 1 sample per input and about 3170x at 100 samples per input, using the authors' released latency script (repo hpankka/revbrain, commit 4418b775) on one Kaggle T4 GPU. Bounds are half the paper's ratios (26 and 1585); 10 invocation pairs; percentile bootstrap. Contains the decision code, harness, power simulation, planted-effect table and the run kernel. Deposited before the main run. Outcome wording: Replicated / Did not replicate / Inconclusive under protocol X.