A comparison of a naive RAG pipeline against an upgraded "context engineering" pipeline shows that the naive version fails at each of four bricks—document parsing, question parsing, retrieval, and generation—producing confident wrong answers. The upgraded pipeline, pointed at a research paper, a NIST standard, and a report with a broken table of contents, returned a typed, cited answer each time. The naive baseline, a 100-line pipeline that parses the PDF, turns the question into keywords, keeps the top few pages by match, and asks the model to answer, works on short, clean prose papers like "Attention Is All You Need" and the RAG paper. However, the gap opens where each loose brick hands the model the wrong context, which the model then answers faithfully. The article isolates each brick by swapping it for its naive version while leaving others reasonable, demonstrating that the same symptom—a confident wrong answer—arrives four ways, one per brick. The usual reflexes of rewriting the prompt, shrinking chunks, or swapping the embedding model do not help because the wrong answer comes from a brick upstream, not the prompt. Getting all four bricks right is what the series calls context engineering. Every failure in the article is a real run, reproduced in a companion notebook on GitHub at doc-intel/notebooks-vol1.