Plagiarism detectors are widely used, yet it is unclear if they match the ways students actually obfuscate copied code. Existing evaluation datasets are limited, often private, or rely on synthetically generated plagiarism instances. In a collaboration between two universities, we conduct an empirical study that challenges students at both institutions to plagiarize given solutions and document their changes. Using this dataset, we analyze their success in evading detection, present common plagiarism strategies, and provide a reliable resource for future research on academic integrity. While only few submissions plagiarize successfully, our participants incorporate a large variety of creative manual and LLM-based strategies.