| The source material contained two different concepts under the same Bayesian/Probabilistic LangChain umbrella: an SMC tool runtime and a heuristic beam framework (`p_langchain`). The beam framework was not retained as Bayesian code because its softmax score was not a posterior and its multi-proposer executor overwrote earlier proposers. The useful SMC idea is retained in `smc.py` with proposal correction and safe final weighting. The Colab RAG export was replaced by a package-native RAG implementation and reproducible benchmark. |
