Notebook

Working notes, reading logs, and in-progress thinking, newest first. Entries here are exploratory by design; finished work graduates to Research.

1. When the Nearest Neighbor Is Everyone's Neighbor

   Aug 3, 2026

   A 50-document retrieval experiment where cosine similarity's aggregate metrics looked fine while a handful of documents kept surfacing for unrelated queries. Measuring the neighbor graph directly, and correcting it with CSLS, separates a geometric problem from a semantic one, and shows why fixing the first barely touches the second.

   • embeddings
   • cosine-similarity
   • semantic-search
   • retrieval
   • hubness
   • csls
   • rag