A news organization deploys a RAG system over an archive of 10 million articles spanning 20 years. The system is used by journalists to research stories. A journalist asks: 'What is the current policy on immigration detention?' The system retrieves the 5 most semantically relevant articles — which happen to be from 2019 — and Claude provides a confident answer based on outdated policy information. No disclaimer is provided. The retrieval system has no temporal metadata filtering. What architectural combination best addresses knowledge freshness in this production system?
Select an answer to reveal the explanation.
Short Explanation
You need retrieval that favors newer articles and a warning when even the newest source is stale. Recency-weighted retrieval with a present-tense date pre-filter keeps the search current, and the 180-day disclaimer warns journalists before they rely on outdated policy information.
Full Explanation
The best architecture is recency-weighted retrieval with a date pre-filter for present-tense queries plus a mandatory freshness disclaimer when the most recent retrieved document is older than 180 days. Recency weighting ensures the retrieval pipeline favors recent articles for time-sensitive queries, and the present-tense date pre-filter narrows current-state queries to recent material before ranking. But recency weighting alone can fail when a very relevant 2019 article outscores a less semantically similar 2024 article, still retrieving stale content, or when the most recent article about the topic is still from 2022. The freshness disclaimer acts as a mandatory safety net: regardless of what retrieval returns, Claude checks the publication dates of retrieved documents and flags when the freshest source is over 180 days old, giving journalists the critical context that this information may be outdated. The approach that adds publication date metadata to all document chunks and combines vector similarity with recency decay improves retrieval but doesn't prevent stale results from dominating and provides no user-facing signal about staleness. The hard 90-day filter for current queries is too restrictive: for some policy topics, no articles exist from the last 90 days, returning empty results. The citation-date-only approach pushes responsibility entirely to users without improving retrieval, and journalists researching stories may not notice a 2019 date embedded in a long response.