The spec sets 5-8 sub-queries per question. After the first benchmark run, check:
- Are redundant axes wasting Tavily API calls?
- Do some question types need fewer/more sub-queries?
- Is the total cap of 60 pre-dedup results appropriate?
Files: bioscancast/stages/search_stage/query_decomposition.py, pipeline.py
The spec sets 5-8 sub-queries per question. After the first benchmark run, check:
Files:
bioscancast/stages/search_stage/query_decomposition.py,pipeline.py