Eligibility Scores and Reputation

A game-theoretic model of platform scores, reliance, and reputation in digital markets.

Overview

This research project studies platform scores that are not only informative signals, but also rules of access to a digital infrastructure. The motivating example is an advertising platform where a favorable eligibility score increases the exposure of an ad by allowing it to enter more auctions, receive more impressions, or access better allocation channels.

The central concept is reliance: the extent to which an automated system, a marketplace, or an allocation pipeline mechanically relies on a score when making decisions. A score with high reliance can create more value, but it also creates stronger short-term incentives for the platform to be too permissive.

Main idea

The model formalizes a dynamic trade-off. When a platform gives a favorable score to a low-quality object, it obtains an immediate gain from additional exposure. However, if this permissiveness is detected, part of the infrastructure may stop trusting the score. The platform then loses a future franchise value attached to the credibility of the scoring system.

This mechanism can generate a coordination problem. If the score is widely used, the platform has a lot to lose from damaging its credibility, which can discipline its behavior. If the score is barely used, the future loss is small, discipline is weak, and low reliance can become self-fulfilling.

Contributions

• Defines eligibility scores as objects that are both informational signals and infrastructure rules.
• Introduces a reliance map linking score reliability, platform discipline, and future usage of the score.
• Shows how low reliance can become a stable trap, even when a high-trust equilibrium also exists.
• Studies the role of detection effort and transparency in preserving the credibility of the score.
• Provides a numerical illustration of multiple equilibria, bifurcation, hysteresis, and safe online learning.

Methods

The project uses tools from game theory, reputation models, platform economics, and dynamic systems. The formal analysis characterizes incentive constraints, fixed points of the reliance map, stable and unstable equilibria, and a sufficient safety frontier for detection effort. A numerical section illustrates the theoretical mechanism through simulated reliance dynamics.

Current status

This is an ongoing research-oriented project developed during my M1 work. The next step is to make the numerical analysis more systematic, in particular by mapping the robustness of the reliance trap and by developing the online learning extension into a safer exploration problem.