Long-Term Stability of Reputation Aggregation Under Coordinated Attacks

Abstract

Reputation-based trust mechanisms are commonly adopted in decentralized service systems. Their behavior under sustained adversarial conditions, however, remains insufficiently understood. A simulation platform was developed to model 5,000 autonomous agents interacting over dynamic service networks. Three representative aggregation algorithms were evaluated under varying proportions of collusive agents ranging from 5% to 30%. Each experiment was executed for 200,000 interaction rounds. Results show that traditional weighted averaging schemes experience reputation collapse within 40,000 rounds when malicious participation exceeds 15%. By contrast, bounded temporal aggregation maintains stable trust rankings for up to 110,000 rounds under identical conditions. Variance of reputation scores was reduced by 46% on average. The findings suggest that long-term stability should be treated as a primary design objective in trust systems.