The Algebraic Content of Mod 8 Residue Structure in Collatz Stopping Times

We propose a structural origin for the mod-8 residue-class random effect that Bonacorsi and Bordoni (2026) introduce empirically in their hierarchical NB2-GLM for Collatz stopping times. By tracking the symbolic Collatz state through its deterministic prefix, we identify a closed-form per-class structural offset α_det(r) = prefix steps(r) + K_h · log(a_final(r)/2k), where K_h = 3/log(4/3) is the textbook Collatz drift coefficient and a_final is the prefix’s terminal multiplier. We conjecture that the empirical mod-8 random effect equals αdet at k = 3, rescaled by the GLM’s log link with leading-order coefficient 1/E[σ]. Evidence runs on two sides: a 40-cell empirical verification of αdet’s bridge to Tao 2022’s (5.15) at slope unity without parameter tuning, and a head-to-head Bayesian replication under Pathfinder VI at N =107 plus HMC validation at N = 104. The HMC arm confirms the conjecture’s exhaustion content predictively: the mod-8 random effect’s posterior σu vanishes once a_final at k = 6 is in the model. The conjecture is supported but not proven; HMC at production scale is the natural follow-up.