Retrieval Decision Cards
Recall the pre-prompt retrieval checks that prevent context stuffing and stale claims.
Similarity vs fit High similarity or high task fit? Use similarity to find candidates; use task fit to select prompt context. Context stuffing "The model can decide which memories matter if we include them all." A teammate wants to pass the top 20 memory hits into every prompt. Your line Let the retrieval layer filter first. The model should reason over selected evidence, not police access, expiry, and relevance under token pressure. More context can hide the one memory that actually matters. It separates evidence selection from answer generation. Access rule What check happens before a retrieved memory reaches the…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in