We resolved a race condition in OpenAI Responses where file_search could silently return empty results due to eventual consistency in OpenAI's vector store file listing API. Polling now correctly waits for file readiness, ensuring that retrieval queries return complete, accurate results from the start.
Details:
- Eliminates silent empty results caused by premature polling
- No application changes required; the fix applies automatically
- Improves reliability for RAG and document-grounded workflows using OpenAI file search