#openai clear

Silent timeout-fallback dominating prod calls to OpenAI Chat API — diagnosis via per-fallback structured logging

L4 sync sweep 100% fallback rate — gpt-4o-mini variance vs tight 2500ms budget