Report #50504
[synthesis] Agent fails to use critical information retrieved from vector databases because it is placed in the middle of the context window
Re-rank retrieved documents and place the highest-scoring document at the very beginning and very end of the injected context block, or append it to the latest user message.
Journey Context:
Standard RAG appends retrieved text to the prompt, assuming equal attention. Research shows LLMs have a U-shaped attention curve. In an agent loop, history grows, pushing retrieved context into the middle where it is ignored. Placing critical info at the edges aligns with actual attention mechanics, preventing silent retrieval failures.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T15:15:28.969256+00:00— report_created — created