Relative performance evaluation (RPE) is widely used to filter out common shocks, but it is prone to collusion. We study the performance of RPE when agents are differentially productive (or evaluated in a biased manner). While such diversity is always costly in a static setting, it can be useful in repeated interactions, because it makes it harder for agents to collude. We identify conditions under which the principal prefers independent or joint performance evaluation if agents are identical but prefers RPE if they are diverse. In low-skill industries, the principal should offer asymmetric contracts even to homogeneous agents—a form of favoritism—as the cheapest way to combat collusion. Our results generate novel empirical predictions and contribute to the recent literature linking accounting and labor economics.

Form 10-K offers a setting for studying how users process complex, multi-layered disclosures: managerial narratives in the main file alongside separate exhibits, such as contracts and certifications, that provide unfiltered detail. Drawing on rational inattention theory, we investigate how users allocate limited attention across these components. Users typically begin with the main file and selectively access exhibits when the main file appears shorter, less readable, or less confident, indicating higher perceived information loss. This pattern strengthens for exhibits that offer more detail on topics discussed in the main file and among institutional investors and time-constrained users. Exhibit access persists beyond the initial filing window and increases around subsequent firm events, especially when external monitoring strengthens and event-related information asymmetry grows. Collectively, our findings underscore the active, discerning nature of user attention in navigating multi-layered disclosures and reveal the often-overlooked informational value of exhibits in Form 10-K.