Executable documents such as R Markdown combine narrative text, executable code, and rendered outputs in a single source artifact. This approach supports reproducible research and repeatable reporting, but it also creates a wider failure surface than conventional scripts: defects may arise from code logic, data handling, chunk options and execution order, YAML metadata, package dependencies, environment configuration, and the rendering/conversion toolchain. Despite the popularity of R Markdown in data analysis and scientific communication, there is limited empirical evidence on which defect types dominate when code and documents co-evolve. This thesis provides an empirical analysis of defects in R and R Markdown projects by mining GitHub repositories and classifying bug-fix commits using an executable-document–aware taxonomy.