04. Week 8 — Daily Recall¶
Spaced practice. Answer from memory. If stuck, jump to the explainer section shown in parentheses.
Monday (after ELI5 + chapter 1)¶
- In one sentence, why does basic RAG stall around 70% on hard tasks? (§1.1)
- Walk the Q3-vs-Q4 revenue example and say where retrieval fails first. (§1.2)
- Why is “wrong chunk” often worse than “no chunk”? (§1.3)
- Module 07 librarian vs Module 08 head researcher — what changed? (ELI5)
- Name all five placeholders: rewriter, hypothesis, cross-checker, multi-step plan, confidence gate. (ELI5)
Tuesday (after chapter 2)¶
- Rewrite vs expansion — difference in one clean example. (§2.1)
- Decomposition — when is one retrieval call structurally insufficient? (§2.3)
- Step-back prompting — what does it ask for first, and why? (§2.4)
- Take “How did churn change after the pricing update?” and decompose it verbally. (§2.3)
- What is the biggest risk of query rewriting? (§2.1)
- Draw the query transformation flow from memory. (§2.5)
Wednesday (after chapter 3)¶
- HyDE — why embed a hypothetical answer instead of the raw query? (§3.1)
- When does HyDE help most? When can it hurt? (§3.2)
- Parent-child retrieval — what is indexed, and what is returned? (§3.3)
- Hybrid retrieval — why keep both dense and sparse signals? (§3.4)
- What does RRF do, and why is it convenient? (§3.4)
- If the corpus is acronym-heavy, which retrieval upgrade would you try first? (§3.4)
Thursday (after chapter 4)¶
- Retriever vs reranker — what architectural difference creates the quality gap? (§4.1)
- Why rerank top-K rather than the entire corpus? (§4.1)
- Metadata filtering — give three filters that matter in enterprise search. (§4.3)
- MMR — what balance does lambda control? (§4.4)
- Why can top-5 by pure score be worse than a diverse top-5? (§4.4)
- Draw the retrieve → rerank → filter pipeline from memory. (§4.5)
Friday (after chapter 5)¶
- CRAG — what exactly is being corrected? (§5.1)
- Self-RAG — what new behavior appears compared with one-shot retrieval? (§5.2)
- What does the confidence gate inspect before allowing an answer? (§5.2)
- Give one query that should route to decomposition, not plain rewrite. (§5.4)
- Iterative retrieval — what signal tells you to search again? (§5.3)
- Why is this module already “agent thinking,” just inside retrieval? (§5.4)
Weekend (cumulative)¶
- List 8 rows from the failure-fix table in explainer §6.1.
- Explain the full advanced RAG loop in under 60 seconds. (§6.2)
- Answer two interview questions from explainer §6.3 without notes.
- Say one production lesson from §6.4 that you would enforce on a real team.
- Recite the bridge sentence to Module 09 from §6.6.