Retrieval feedback
A recorded signal evaluating the utility of a specific retrieval result. Used to improve future retrieval weighting and reinforcement priority. Stored in the retrieval_feedback OpenSearch index.
Retrieval feedback is what allows the memory gateway to improve over time. After a cognitive loop iteration completes and the outcome is evaluated, the system looks back at which retrieved memories actually influenced the deliberation and how. A memory that was retrieved, cited in the planner's proposal, supported by the critic, and appeared in the winning arbitration decision receives high retrieval utility feedback. A memory that was retrieved but ignored by all agents - never referenced in any proposal - receives low utility feedback. This signal is stored in the retrieval_feedback OpenSearch index and used in two ways: it adjusts the individual memory's retrieval priority (memories that consistently prove useful when retrieved surface more readily in future queries), and it feeds the memory gateway's retrieval strategy tuning (if a specific retrieval configuration consistently surfaces low-utility memories, the weighting between BM25 and k-NN, or the depth parameters, can be adjusted). Retrieval feedback is the mechanism that makes the retrieval pipeline a learning system rather than a static ranker.