Temporal Counting in MaleCNS-Derived Artificial Recurrent Networks: An Independent Replication, Memory-Capacity Extension, and Synchronized Blackjack Simulation

Publication date: 2026-09-13

Contributors: Ogdn Ames

Abstract

Structural connectomes constrain anatomical topology but do not specify functional neural dynamics. An initial MaleCNS v1.0 study found that a primary fixed reservoir lost to random graphs, whereas a postsynaptic-normalized comparison favored native topology on average across only three computational seeds. We conducted a prospectively specified independent replication using ten new input/readout/null seeds on the same 5,000-neuron anatomical subset, 2,048 fresh training shoes, 512 validation shoes, and 1,024 locked test shoes. Native, random, and degree/raw-strength-preserving families received identical training budgets. Fixed baseline true-count root-mean-square error (RMSE) averaged 0.8085 for native models and 0.9990 for random models. The primary seed-generalized comparison favored native topology. A validation-only search over longer-memory linear dynamics and 64, 128, or 256 readout states yielded mean RMSEs of 0.0097, 0.0150, and 0.0190 for native, random, and raw-strength-null families. An original GRU transferred to fresh shoes achieved 0.0928; exact accumulation remained error-free. A separate paired simulation covered 4,096 fresh shoes and 173,440 initial rounds. No estimator-versus-exact paired interval excluded zero. A synchronized video maps frozen artificial states to real soma coordinates while showing exposed cards, count errors, and same-stream settled equity. Results concern constrained artificial computation, not biological counting or profitability. Single-specimen anatomy, effective signed-strength confounds, and restricted dynamics limit generalization.

Keywords: connectomics, reservoir computing, computational replication, temporal memory, blackjack simulation

Published by Ames Investment Systems

License: All rights reserved
Permission is required for reuse beyond applicable legal exceptions.