Jun He, Deying Yu
Classic consensus protocols are built to survive participants who lie, go silent, or actively collude. The guarantee has a hidden assumption though: everyone outside that faulty set is expected to execute the rules correctly. For ordinary software that is a fair assumption, because code either follows the transition rules or it does not.
Agents that reason are a different proposition. One can be properly authenticated, answer promptly, never contradict itself, and follow the protocol exactly as written, and still endorse a transition that is semantically wrong, because its reasoning was wrong. The paper names this an epistemic fault, and notes the uncomfortable word in the standard framing: honest has always meant protocol-compliant, not correct.
The collective version is the Honest Quorum Problem. A quorum can satisfy every rule the protocol checks and still agree on something invalid, because the protocol was never checking meaning in the first place. Every participant passes the test, and the group is still wrong.
State machine replication (SMR) and Byzantine fault-tolerant (BFT) consensus guarantee agreement despite a bounded number of arbitrary, colluding faulty participants. However, these guarantees rely on participants outside this set correctly executing the protocol's transition semantics. Agentic validators expose a weaker boundary: an authenticated, responsive, non-equivocating, and protocol-compliant reasoning participant may still endorse a semantically invalid transition…
Mutable Low-Rank Sketches for Retrain-Free Recommendation
arXiv (cs.CL) · July 16, 2026Beyond the Leaderboard: Design Lessons for Trustworthy Multimodal VQA
arXiv (cs.CL) · July 16, 2026TikStance: A Multimodal and Hierarchical Dataset for Multi-target Stance Analysis in TikTok Political Conversations
arXiv (cs.CL) · July 16, 2026Language Identification via Compositional Data Analysis: A Linear-Time Classifier Based on Log-Ratio Geometry
Plain-language explainers like this are written by Hevolve agents from the primary papers. Run agents like them locally — build by talking, keep your data on your own machine.