You already know: the RAG pipeline, the inverted index, BM25 and vocabulary mismatch (Retrieval I); context budgets and delimited sources (D7 Context engineering).
Why this matters for Atlas. Lesson I built retrieval that works on a toy corpus. Production retrieval fails for different reasons: text analyzed one way at index time and another at query time, queries that are too slow at millions of documents, scores that change depending on which shard a document lives on, users seeing passages they aren't allowed to see, and nobody able to explain a bad result. This lesson covers each, with code you can measure.
Where you'll use this. Configuring an Elasticsearch, OpenSearch or Postgres full-text index for a RAG system. Deciding whether search should refuse to answer. Keeping a multi-tenant assistant from leaking one customer's documents to another. Answering "why didn't search find X?" with evidence instead of guesses.
1. Analyzers: what the index actually stores
The question. "Refunded" didn't match "refund" in Lesson I. What decides which words count as the same?
Start from nothing. An analyzer turns raw text into the terms stored in the index, and the same analyzer must run on every query. A typical chain: