Indexing a document is easy. Keeping millions of vectors in sync with it is the job.

Day one of a RAG system is chunk, embed, insert. Then people start editing documents. Each fix below solves one production problem and exposes the next. Every number on this page is computed live from a simulated index, so you can replay any step.

Simulation: 121 fictional documents, feature-hashed bag-of-words embeddings (two seeds = two incompatible models), an in-memory vector store with an async upsert queue. It illustrates the mechanics; it is not a benchmark of any vector database. Source: src/utils/prodLab.js.