Phase 2: statscore-bayesian — conjugate models, MCMC + Python

[armanrasta]

Summary

Bayesian inference: conjugate posteriors and MCMC samplers. Python bindings ship alongside.

Rust (statscore-bayesian)

• Conjugate: Beta-Binomial, Normal-Normal, Gamma-Poisson, Dirichlet-Multinomial
• MCMC: Metropolis-Hastings, Gibbs
• Diagnostics: Geweke, Gelman-Rubin

Python bindings (statscore.bayesian)

• Posterior sampling returns NumPy arrays
• Trace plot helpers (optional matplotlib)

Done when

• Conjugate posteriors match analytical solutions
• MH sampler passes toy distribution tests
• Python API documented