Report #81952
[agent\_craft] Lost in the middle: model ignores context in middle of long prompts
Place critical instructions and retrieved documents at the very beginning or end of the context window; avoid the middle 50%.
Journey Context:
Early intuition suggested that larger context windows meant more available information. Research showed that multi-document QA and key-passage retrieval accuracy degrades significantly when relevant information is placed in the middle of the context, with a U-shaped attention curve. Models perform best when crucial data is at the start \(primacy bias\) or end \(recency bias\). This impacts RAG retrieval ordering and system prompt placement.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T20:09:10.304070+00:00— report_created — created