Research
feed-research
Teardowns of big-tech feed algorithms. Primary sources only, and every claim carries an evidence grade (verified observed inferred). A batch job reviews newly published material every night at 01:00 KST. These pages are machine translations (Claude) of the Korean originals, checked automatically for links, numbers, evidence grades and code. As of 2026-11-10.
Daily reports
What we read each day and how our understanding changed. The canon holds the conclusions; these reports explain them.
Service canon
- FacebookThe latest public canon for Facebook Feed ranking is the 2021 structure: a friend-graph lookup, candidate redu…Updated 2026-10-22
- InstagramInstagram documented Explore, Feed and Reels Chaining as separate AI systems, and as of the system cards read …Updated 2026-10-22
- NetflixNetflix recommendation is a system that optimizes page (row × column) composition over a curated catalog (thou…Updated 2026-10-19
- RedditAs of the 2026-10-28 revision, the Reddit help page describes home feed recommendation in four stages: candida…Updated 2026-11-01
- ThreadsThreads is an app built in 5 months on Instagram infrastructure (infrastructure section). Yet its ranking move…Updated 2026-10-22
- X (Twitter)As of the 2026-11-05 commit (021afa4, Gen4+57), X For You gathers the latest in-network posts from Thunder and…Updated 2026-11-08
Cross-service topics
- Candidate allocationCandidate allocation is the layer that decides, when there are 10 or 30 candidate generati…
- Embedding-based candidate generationCandidate generation (retrieval) is the stage that first picks, from all content, the cand…
- Feed diversitySorting only by ranker score produces runs of the same author, the same topic, and the sam…
- LLMs in the recommendation pipelineBased only on primary material released through 2026-08-13, X put an LLM in as a judge, Me…
- Model freshnessMeta calls the speed from collecting a signal to training on it and using it for recommend…
- Multi-task rankingMulti-task ranking is a structure in which the ranker does not predict "will this user lik…
- Sequence Transformer ranking (in-context candidate scoring)Sequence Transformer ranking puts a user's behavior history in as a token sequence, append…
- Survey signalsEngagement logs tell us only what users did. A survey asks users directly. It asks "Was th…