History — 2026 week 24 (Jun 8 – Jun 14, 2026)
Newest first.
2026-06-13 — Elasticsearch In-Engine LTR (~6 notes)
Filled the Elasticsearch-internal side of the LTR cluster, complementing the existing external/Metarank coverage.
Concepts — Elasticsearch Learning to Rank — hub note covering two distinctions: external re-ranker (Metarank) vs. in-engine LTR, and the two ES modules — the OpenSource Connections plugin (RankLib-based, 2017) vs. Elastic native LTR (8.12+, GBDT/XGBoost via eland).
Articles — We’re Bringing Learning to Rank to Elasticsearch (by OpenSource Connections / Doug Turnbull, 2017-02-14). Launch of the o19s elasticsearch-learning-to-rank plugin; integrates RankLib; custom ltr query + ranklib script; applied via rescore. Learning To Rank (LTR) - Elasticsearch Native (Elastic docs). Native LTR overview — judgment lists, three feature categories, query_extractor templated features, GBDT/LambdaMART inference with training outside ES. Search using LTR - Elastic Docs (Elastic docs). Applying a deployed model as a learning_to_rank rescorer (8.12+) or rescorer retriever (9.1+); mandatory window_size.
Tools — eland — Elastic’s Python client; uploads the trained XGBoost ranker into Elasticsearch for native LTR inference.
Updated — Learning to Rank, Metarank, Elasticsearch, RankLib, OpenSource Connections, MOC - Ranking and Retrieval cross-linked to the new cluster.
2026-06-12 — PostgreSQL Search Extensions (~7 notes)
Concepts — Search Results Explainability (ability to reason about and explain why documents rank as they do), Learned Sparse Retrieval (family of models learning term weights for sparse retrieval — SPLADE, ELSER, etc.).
Tools — pg_trgm (PostgreSQL built-in trigram similarity extension for fuzzy text search), pg_textsearch (BM25 for PostgreSQL by Tiger Data), pgvectorscale (high-performance ANN vector search extension for PostgreSQL by Tiger Data), VectorChord (PostgreSQL vector search extension by TensorChord).
Companies — Tiger Data (formerly Timescale; builds PostgreSQL-native search and vector search extensions).
2026-06-11 — Metarank / LTR Cluster (~8 notes)
Articles — Hybrid Search and Learning-to-Rank with Metarank (by Vsevolod Goloviznin, Pinecone blog, 2023-06-30). Fusing BM25 + vector retrievers via LambdaMART as a statistically-valid fusion mechanism; cold-start via Interleaving. Learn-to-Rank with OpenSearch and Metarank (by Roman Grebennikov, OpenSearch blog, 2022-10-25). LambdaMART secondary re-ranking on top of OpenSearch; multi-stage ranking pattern; Remote Ranker Plugin RFC. Metarank - Personalized Ranking That Actually Reads Your Clicks (by Florian Narr, Codeline, 2025-12-15). Code-level repo review of Metarank’s Scala/fs2 architecture (FeatureMapping, MetarankFlow, RankApi). OpenSearch vs. Elasticsearch in 2025 - What’s Changed and What Hasn’t (Dattell, 2025-04-03). Balanced licensing/features comparison (SSPL vs. Apache 2.0, plugin incompatibility, performance differences, ecosystem divergence).
Tools — Metarank — open-source LambdaMART re-ranker and personalization service (Scala + Redis) by Roman Grebennikov and Vsevolod Goloviznin.
People — Roman Grebennikov (Metarank co-creator), Vsevolod Goloviznin (Metarank co-creator, Pinecone author), Florian Narr (Codeline author).