Memetic Collapse-Chain Optimisation (MCCO) is a five-layer self-adaptive memetic algorithm specifically designed for sequential decision under uncertainty. We are not aware of any prior memetic framework that exploits closed-form Bayesian state-update operators with provable contraction properties for state-conditioned local search. The central conceptual contribution is structured semantic diversity: rather than relying on random population initialisation as in conventional memetic algorithms, MCCO derives its population from multiple structurally distinct observables applied to the same raw data. The five layers operate at distinct temporal scales: (1) single-chain BPTT for gradient-based parameter refinement, (2) multi-path Monte Carlo gradient averaging across stochastic futures, (3) a Chain Matrix of parallel observation chains with genetic algorithm population management, (4) cross-day memory with best-ever permanent seeding and age-limited elite retention—elevated to a generalisable Explore-Without-Regress design principle—and (5) meta-learning that auto-adjusts hyperparameters. For a representative industrial GA+MC configuration, MCCO reduces operational complexity from ~3. 6×10⁸ to ~1. 5×10⁴—an illustrative ~24, 000× speedup, configuration-specific rather than universal. Concept verification on a 7-day H1 financial parameter optimisation case study demonstrates functional behaviour across all five layers in 65 seconds on consumer hardware. A landscape-guided deployment variant achieves an additional 3. 56× speedup on unimodal loss surfaces.
H Y Rao (Sun,) studied this question.