Credibility Governance: A Social Mechanism for Collective Self-Correction under Weak Truth Signals

Opinion aggregation on social media and web platforms increasingly determines how real-world resources are allocated. Yet current aggregation methods rely on easily amplified indicators such as vote-based engagement or capital-weighted commitments. These signals reflect visibility rather than reliability, making collective judgments vulnerable to early surges, strategic manipulation, and uneven evidence, especially when truth signals are weak, noisy, or delayed. To address this issue, we propose Credibility Governance (CG), a mechanism that reallocates influence based on how well agents track evolving signals rather than on voting or capital levels alone. CG assigns each opinion and each agent a dynamic influence score (credibility) that reflects how trustworthy they appear under observed signals. It updates opinion influence through weighted endorsements from reliable agents, and reciprocally updates agent influence based on the long-term performance of the opinions they support, rewarding early insights and consistent alignment with emerging evidence. This enables the system to distinguish persistent trends from short-lived noise. We evaluate CG in POLIS, a socio-physical simulation environment that models the co-evolution of social beliefs and physical feedback under uncertainty. Under weak signals, noisy observations, and initial majority misalignment, CG surpasses vote-based, capital-weighted, and no-governance baselines by enabling (i) earlier and more stable recovery of the true state, (ii) stronger resilience to misinformation shocks, and (iii) sustained support for correct minority viewpoints. By growing influence through demonstrated correctness rather than surface-level signals, CG contributes a governance mechanism that improves collective epistemic quality and supports more trustworthy, equitable, and socially beneficial online ecosystems.