Pre-registration (frozen 4 Nov 2026): replication of the GatedV value-gating perplexity improvement (arXiv 2611.02145), as published and under a matched learning-rate search

Frozen protocol, decision code, planted-effect table and kernel for a private pre-registered replication, deposited before any run. Clauses: mean paired held-out perplexity change below -1.0% (half the paper's -2.03%) and at least 8 of 10 pairs improving, for the as-published arm and, separately, a matched arm with a three-rate learning-rate search for both models. Outcome wording: Replicated / Did not replicate / Inconclusive under protocol X.