fix: increase top_k to 10 and remove last [N] from verification prompt

Two fixes:
1. Increased default top_k from 5 to 10. With top_k=5, short queries
like "China in 2024?" missed the relevant page (e.g., page 410)
because generic intro pages ranked higher. More chunks = better
chance of retrieving the right content.

2. Removed the last literal [N] from the verification prompt template
(checklist item 1) to prevent any residual [N] in output.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

README.md

@@ -346,7 +346,7 @@ embeddings:
 retrieval:
   chunk_size: 1000                 # Characters per chunk (when splitting documents)
   chunk_overlap: 100               # Overlap between chunks (preserves context)
-  top_k: 5                         # Number of chunks to retrieve per query
+  top_k: 10                        # Number of chunks to retrieve per query
 
 web_search:
   enabled: true                    # true | false

setup.py

@@ -48,7 +48,7 @@ def generate_config(
         "retrieval": {
             "chunk_size": 1000,
             "chunk_overlap": 100,
-            "top_k": 5,
+            "top_k": 10,
         },
         "web_search": {
             "enabled": web_search,

src/prompts.py

@@ -94,7 +94,7 @@ problems.
 
 Run through this 9-point verification checklist and report results:
 
-1. Does every factual claim have a citation [N] endnote?
+1. Does every factual claim have a numbered citation endnote (e.g. [1], [2])?
 2. Are all citations grounded in the provided context?
 3. Are there any claims that appear fabricated or hallucinated?
 4. Are direct quotes accurately reproduced from the context?

src/retriever.py

@@ -14,7 +14,7 @@ def retrieve_from_vectordb(query: str, cfg: dict) -> list[dict]:
     """Retrieve relevant chunks from ChromaDB."""
     from src.ingest import get_chroma_collection
 
-    top_k = cfg.get("retrieval", {}).get("top_k", 5)
+    top_k = cfg.get("retrieval", {}).get("top_k", 10)
     collection = get_chroma_collection(cfg)
 
     if collection.count() == 0: