Clarify refine_factor, set default to 30

CHANGELOG.md

@@ -7,7 +7,7 @@
   - Creates IVF_PQ indexes
   - Requires minimum 256 chunks (LanceDB training data requirement)
   - New `search.vector_index_metric` config option: `cosine` (default), `l2`, or `dot`
-  - New `search.vector_refine_factor` config option (default: 10) for accuracy/speed tradeoff
+  - New `search.vector_refine_factor` config option (default: 30) for accuracy/speed tradeoff
   - Indexes not created automatically during ingestion to avoid performance degradation
   - Manual rebuilding required after adding significant new data
 - **Enhanced Info Command**: `haiku-rag info` now shows storage sizes and vector index statistics
@@ -17,7 +17,7 @@
 
 ### Changed
 
-- **Evaluations**: Improved evaluation dataset naming and simplified evaluator 
+- **Evaluations**: Improved evaluation dataset naming and simplified evaluator
 - configuration
   - `EvalDataset` now accepts dataset name for better organization in Logfire
   - Added `--name` CLI parameter to override evaluation run names

docs/configuration.md

@@ -88,7 +88,7 @@ research:
 
 search:
   vector_index_metric: cosine  # cosine, l2, or dot
-  vector_refine_factor: 10
+  vector_refine_factor: 30
 
 agui:
   host: "0.0.0.0"
@@ -732,14 +732,15 @@ Configure vector indexing behavior for efficient similarity search:
 ```yaml
 search:
   vector_index_metric: cosine  # cosine, l2, or dot
-  vector_refine_factor: 10     # Re-ranking factor for accuracy
+  vector_refine_factor: 30     # Re-ranking factor for accuracy
 ```
 
 - **vector_index_metric**: Distance metric for vector similarity:
   - `cosine`: Cosine similarity (default, best for most embeddings)
   - `l2`: Euclidean distance
   - `dot`: Dot product similarity
-- **vector_refine_factor**: Retrieve `refine_factor * limit` candidates and re-rank in memory for better accuracy. Higher values increase accuracy but slow down queries. Default: 10
+- **vector_refine_factor**: Improves accuracy when using a vector index by retrieving `refine_factor * limit` candidates (using approximate search) and re-ranking them with exact distances. Higher values increase accuracy but slow down queries. Default: 30
+  - **Only applies with a vector index** - has no effect on brute-force search, which already returns exact results
 
 !!! note
     Vector indexes are only necessary for large datasets with over 100,000 chunks. For smaller datasets, LanceDB's brute-force kNN search provides exact results with good performance. Only create an index if you notice search performance degradation on large datasets.

haiku_rag_slim/haiku/rag/config/models.py

@@ -83,7 +83,7 @@ class ProcessingConfig(BaseModel):
 
 class SearchConfig(BaseModel):
     vector_index_metric: Literal["cosine", "l2", "dot"] = "cosine"
-    vector_refine_factor: int = 10
+    vector_refine_factor: int = 30
 
 
 class OllamaConfig(BaseModel):