Answers are grounded in this paper and cite their sources.
User feedback can help agricultural question-answering services adapt. Storing an unverified correction can also carry an error into later answers. We examined these two uses of feedback separately using 160 judgments from eight participants, controlled simulations, and exploratory language-model tests. In simulation, the full router combined reliability weighting with replay and showed no statistically distinguishable advantage over equal weighting. In the local-model test, unfiltered corrections changed 16 of 78 responses from matching the original key to disagreeing with it. Alternative reference judgments changed the flip counts, and the server-model results also depended on how agreement was scored. Workflow tests blocked stale updates and restored an earlier snapshot, while erroneous approvals could still enter memory. The results support separating feedback used for response selection from the verification required before a correction is published. A performance advantage for the full framework and safety in agricultural practice remain unestablished.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: