Do Agents Repair When Challenged -- or Just Reply? Challenge, Repair, and Public Correction in a Deployed Agent Forum

Abstract

As large language model (LLM) agents are deployed in public interactive settings, a key question is whether their communities can sustain challenge, repair, and public correction, or merely produce norm-like language. We compare Moltbook, a live deployed agent forum, with five matched Reddit communities by tracing a three-step mechanism: whether discussions create threaded exchange, whether challenges elicit a response, and whether correction becomes visible to the wider thread. Relative to Reddit, Moltbook discussions are roughly ten times less threaded, leaving far fewer chances for challenge and response. When challenges do occur, the original author almost never returns (1.2% vs. 40.9% on Reddit), multi-turn continuation is nearly absent (0.1% vs. 38.5%), and we detect no repairs under a shared conservative protocol. A non-challenge baseline within Reddit suggests this gap is linked to challenge, not simply deeper threading. These results indicate that social alignment depends not only on producing norm-aware language, but on sustaining the interactional processes through which communities teach, enforce, and revise norms. This matters for safety, because correction is increasingly decentralized, and for fairness, because communities differ in how they expect participants to engage with challenge.

Figures (7)

• Figure 1: Three-step mechanism chain. Nested threading (H1) creates opportunities for interaction. Repair (H2) tests whether followup and repair emerge. Public correction (H3) tests whether multi-turn correction becomes visible.
• Figure 2: Cross-platform results across the three-step mechanism chain. Large dots show platform means; small dots show individual community pairs. The gap compounds at every step: flat structure (H1) $\rightarrow$ absent followup and repair (H2) $\rightarrow$ missing public correction (H3). Moltbook values are near zero across all metrics and communities. Cross-platform gaps in followup, original-author return, and multi-turn rate are all significant at $p<0.001$ (permutation test).
• Figure H1: Challenge vs. non-challenge subtrees within Reddit. Challenge-anchored subtrees show higher original-author return and multi-turn rates than ordinary reply-anchored subtrees across the matched communities.
• Figure L1: H3 post-challenge subtree metrics across the five matched pairs. Reddit challenge episodes are far more likely than Moltbook episodes to produce original-author return (left) and multi-turn continuation (right); Moltbook values are near zero across all communities.
• Figure L2: Mechanism chain across the five matched community pairs. Left: H1 structural interaction gap (nesting rate). Center: H2 repair deficit (challenge followup rate). Right: H3 original-author return rate after challenge. The gap compounds across steps: flat structure $\rightarrow$ absent repair $\rightarrow$ weak public correction.