EmbeddedKittens: An Evaluation of Code Embeddings for Scratch

Artefacts for our ACM TOSEM submission ‘EmbeddedKittens: An Evaluation of Code Embeddings for Scratch’. (Preprint: doi:10.48550/arXiv.2607.19291) Contents: sprite_naming_projects.parquet: The Scratch projects used for the sprite naming task. Contains the project JSON, the project ID, and author information to give sufficient credit. Projects originally published under CC-BY-SA-2.0 (projects were created before the change of the terms of service and corresponding change of licence in January 2026). per-language-sprites-RQ2.zip: The Scratch projects used for the language-specific datasets. Same licence terms as for `sprite_naming_projects.parquet` apply. model-implementations.zip: Our implementations of the code embedding models. model-checkpoints-rq1.zip: The model checkpoints after training on the sprite naming task. raw-sprite-naming-results_RQ1_RQ2.zip: The raw results for the sprite naming task. evaluation-scripts.zip: Evaluation scripts for all RQs and evaluation data for RQs 3–5.