Towards Mitigating Fabricated Consensus: The Active Provenance Gate for Multi-Agent Debate Synthesis
The paper proposes a gate that blocks unsupported debate summaries instead of polishing them into false consensus.
The Active Provenance Gate audits post-debate claims against the debate log, applies self-correction, and withholds claims that cannot be traced. In crisis simulations, its self-healing step more than doubled average Provenance Fidelity in difficult conditions before the strict gate produced divergence reports. In a human study, over 75% of users preferred an explicit failure report in critical scenarios, even when the baseline’s fabricated consensus read more fluently. ArXiv · AI/CL/LG's note
The Active Provenance Gate audits post-debate claims against the debate log, applies self-correction, and withholds claims that cannot be traced. In crisis simulations, its self-healing step more than doubled average Provenance Fidelity in difficult conditions before the strict gate produced divergence reports. In a human study, over 75% of users preferred an explicit failure report in critical scenarios, even when the baseline’s fabricated consensus read more fluently. ArXiv · AI/CL/LG's note
score 4