Report #67683
[architecture] Retrieved memories override current context and pollute new answers
Apply a confidence gate or cross-encoder reranker before injecting retrieved memories into the prompt. If the retrieval score is below a threshold, or if the memory directly contradicts an explicit current user message, omit it from context.
Journey Context:
Naive RAG pipelines inject the top-K results directly into the system prompt. If the user says 'I actually hate Python now' but the top retrieved memory is 'User loves Python', the agent gets confused and often defaults to the older, strongly weighted memory. Reranking or gating ensures only highly relevant, non-contradictory memories make it into the precious context window, preventing stale context from hijacking the generation.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T20:05:18.841371+00:00— report_created — created