History — 2026 week 28 (Jul 6 – Jul 12, 2026)
Newest first.
2026-07-12 — Grebennikov MICES 2019 real-time personalization talk (2 notes)
Processed the MICES 2019 talk Roman Grebennikov - Personalizing Search Results in Real-Time (Findify) from YouTube captions — real-time LTR across ~1,500 merchants: Position Bias feedback-loop degradation, the shuffled ~1% exploration segment, one generic cross-merchant model with feature scaling, and purchase-weighted perfect rankings for NDCG (the Stanley bong story). Includes a clickable Key Moments timestamp table derived from caption timings (the video’s official chapters cover only the first half). Also created Exploration vs Exploitation (Concepts → Behavioral Signals & Bias). Backlinked from Roman Grebennikov, MICES, Learning to Rank, LambdaMART, NDCG, Position Bias, Personalization in Search, E-commerce Search; registered in global_toc, Videos, index.md, Index, and All about Information Retrieval & Search. (A Findify company note was deliberately not kept — mentions stay as plain text.)
2026-07-06 — Federated vs Unified Search topic (1 note)
Created Federated vs Unified Search, a Topic hub anchoring a comparison the vault previously held only as two separate concepts: Federated Search (query-time fan-out + merge) vs Unified Search Index (index-time consolidation). Framed around one question — where does integration happen, query time or index time? — with a trade-off analysis (freshness / ranking consistency / latency / governance) and a terminology trap section disambiguating three senses of “federated”: classic distributed IR, the Algolia/e-commerce multi-index UI (a unified engine with federated presentation), and Netflix’s federated data graph (made searchable by indexing into a unified index). Documents the convergence pattern — unified core with federated edges — grounded in vault case studies (Canva - Search Pipeline Modernization, Reddit - Vector Database Selection, Netflix - Content Search Architecture, Bonsai - Designing Search for a Relational Database). Backlinked from both concept notes and MOC - Architecture and Search Team (System Architecture); registered in global_toc (Topics → Domains & Platforms), index.md, Index, and All about Information Retrieval & Search.