Sequential Monte Carlo

Likelihood tempering plus resampling empirically gave an increasingly accurate posterior approximation in the controlled experiment. Ideal TGD has a consistency guarantee for bounded measurable test functions under exact reconstruction kernels and standard assumptions. TGD is formulated as an annealed sequential Monte Carlo sampler. Each TGD stage reconstructs clean candidates, reweights particles by an incremental likelihood ratio, optionally resamples, and re-noises retained particles. The consistency guarantee does not apply to approximate neural reconstruction modules or accelerated pruning.