A perennial problem in executive politics is that agents charged with carrying out tasks have private information about their performance and corresponding incentives to manipulate this information. Conventional wisdom emphasizes outside stakeholders as a source of reliable information. We instead focus on agents with competing interests within the state apparatus. When officials are subject to a relative performance evaluation metric, governing outcomes that are directly observable by others are more likely to be truthfully reported—while others are likely to be misreported. Exploiting a difference-in-differences design with an original dataset on workplace accidents in China, we find that incorporating a “death cap” into the official metric management system has heterogenous effects on reported accidents. The aggregated casualties and accident counts dropped by 33.8% and 29.4% respectively, but the decline was entirely driven by sectors in which peer reporting was not feasible. The results highlight the conditions under which agents are restrained by their peers among other actors. In contrast to theories that truthful signals occur when interests are aligned or independent oversight is present, intra-agent competition can get lower-level agents to reveal more truthful information when alternative mechanisms are weak.