Build a tiny end-to-end RAG in pure stdlib: `mini_rag(query, docs)`. 1. **Embed**: use `fake_embed(s) = sum(ord(c) for c in s) % 100`. (Real embeddings are 1024-dim — this 0-99 scalar is just for the recap.) 2. **Score**: compute `abs(fake_embed(doc) - fake_embed(query))` for each doc. 3. **Retrieve top-3**: sort ascending by distance (tie-break alphabetically to stay deterministic). 4. **Augment**: return `{"query": query, "context": [top-3 doc strings], "answer": f"Based on {len(context)} docs..."}`. The harness prints the query, then each retrieved doc on its own line, then the answer.