Order-book / AMM / hybrid venue simulator

Agent-based simulation framework comparing AMM, CLOB, and hybrid (CLOB + passive AMM layer) derivatives venue designs under event-driven Bayesian trader populations.

Continuous-price log-linear dynamics, multi-market cluster with shared latent factor loadings, four agent classes (naive Gaussian, tail-aware base-rate priors, cross-market aggregated-evidence via cosine-similarity loadings, joint-factor information-form posteriors). 900-run sweep across 3 mechanisms × 2 population mixes × 3 capital regimes × 2 signal levels × 25 seeds.

Cleanest finding: the hybrid mechanism halves agent-to-agent trade volume across every slice because the LP layer absorbs noise flow that would otherwise force informed traders into direct adverse selection. Rent flow turned out to be regime-dependent rather than mechanism-dependent — a null result against prior LS-LMSR work where capital was the binding constraint in tail regimes.