V4 proposal pipeline generates 0 LLM proposals — falls back to random #6
Description
Problem
During 1-experiment test, all research rounds produced 0 valid proposals. Circuit breaker fell back to random.
Fixed (8c7bed4)
- Strip Qwen
<think>tags before parsing - 3-strategy fallback parser (exact, markdown, assignment)
- Few-shot examples + XML sections in prompt
- Diagnostic logging for every rejection
Still Open
- Scholar API rate limiting (429s)
- Full 64-experiment quality validation