0-2-2-understand-colbert-approximate-search

name	0.2.2-understand-colbert_approximate_search
description	[0.2.2] Approximate nearest neighbor indexing on per-token document embeddings enabling efficient retrieval of candida

understand-colbert_approximate_search

CALL NUMBER: deep_dense_retrieval.colbert_approximate_search : retrieval_augmented_generation_architecture_patt(5) DEFINITION: Approximate nearest neighbor indexing on per-token document embeddings enabling efficient retrieval of candidate documents before late interaction re-scoring.

Invoke this skill to understand colbert_approximate_search down to its primitives. The RELATIVE ROOT below is the least-fixed-point closure of everything it bundles from — the full import cone, grouped by the lib each prim comes from. Projected from a prover-typed KB (MAP/SWI-Prolog consistency gate): every reference below resolves.

THE RELATIVE ROOT (the import cone, by lib)

from deep_dense_retrieval

• colbert_document_encoder (d1): Encoder producing per-token document embeddings from document passages, typically run offline and indexed for fast retrieval against query embeddings.
• colbert_representation_rank (d2): Rank dimension of per-token embeddings determining expressiveness of token-level similarity computation in late interaction framework.

from retrieval_augmented_generation_architecture_patt

• approximate_nearest_neighbor (d1): Vector indexing technique trading exactness for speed: HNSW, IVF, or LSH structures enabling sub-linear retrieval at scale
• hybrid_search (d1): Retrieval combining multiple search paradigms (dense/sparse, vector/keyword) with score normalization or late fusion
• colbert_style_retrieval (d2): Late interaction retrieval: compute token-level similarity then aggregate for fine-grained relevance without full cross-encoding
• retrieval_fusion (d2): Combining results from multiple retrieval methods through Reciprocal Rank Fusion, Score Normalization, or learned weights
• reciprocal_rank_fusion (d3): Fusing ranked retrieval lists from multiple systems by reciprocal rank weighting to produce unified ranking

CONSUMERS (what needs this)

colbert_re_reranking

---

Projected from the retrieval augmented generation architecture patterns KB (207 concepts / 225 relations) — consistency-typed by MAP; the facet list after the colon IS the cross-lib dependency web.

(leaf — this is an actual skill.)