Exposure Concentration Under a Learned Ranker Is a Cold-Start Effect: A Convergence Test in a Simulated Matching Market

This paper investigates how recommender-mediated visibility affects two-sided matching markets using an agent-based simulation of 1,000 agents. By comparing a neural collaborative filter against an oracle and exploring horizons up to 300 epochs, the study demonstrates that exposure concentration is a transient cold-start effect rather than a permanent structural floor. The work further evaluates an institutional comparison featuring signaling costs, tests robustness across a twelve-cell parameter sweep, and provides methodological checks against closed-form matching results.