This paper presents a synthetic dataset and evaluation framework for analyzing investment portfolio similarity. Understanding how portfolios relate in terms of structure and performance behavior—such as risk, return, and sensitivity to market movements—is essential for strategy comparison, benchmarking, and risk-based clustering in both research and industry. We generate 1,000 portfolios using a diverse asset universe, compute a rich set of performance and risk metrics, and represent portfolio relationships through similarity graphs. Four similarity functions—Jaccard, Cosine, Euclidean, and Pearson correlation—capture different dimensions of portfolio composition and performance behavior. Using hierarchical clustering and standard evaluation metrics, we show that behavior-based similarities, particularly cosine similarity, produce the most interpretable clusters. Our contributions include a reproducible methodology for synthetic portfolio generation, a set of similarity metrics, and a framework for evaluating clustering effectiveness. This resource supports further research in portfolio modeling, recommendation systems, and financial strategy design.