Patching in esports presents critical technological and managerial challenges, requiring developers to balance novelty, accessibility, and competitive integrity. Prior research has largely overlooked how patching strategies influence sustained player engagement. This study examines which types and complexities of patch changes best support long-term player retention in the esports ecosystem. Using Dota 2 as a case study, we categorized patches from 2012 to 2024 via ChatGPT-4o-mini and constructed time series data linking patch types to player counts. Applying Interrupted Time Series (ITS) and Generalized Least Squares (GLS) models, we evaluated short-term excitement and long-term retention effects, controlling for seasonality and autocorrelation. Findings reveal that while complex patches boost immediate engagement, they risk undermining long-term retention. Our study advances theory on digital ecosystem dynamics and offers practical insights for optimizing patch strategies in esports.