Operations on ranked lists: fusion (RRF, Copeland, CombMNZ, DBSF, 14 methods), reranking (MaxSim/ColBERT, MMR, DPP), evaluation (NDCG, MAP, MRR), diagnostics, pipeline builder.
Embedding interfaces + local backends (Candle/HF).
IR ranking math kernels (no indexing): BM25/TF-IDF/LM transforms.
Learning to Rank: differentiable ranking, LTR losses (RankNet, LambdaRank, ApproxNDCG, ListNet, ListMLE), trainers, and IR evaluation metrics