Google Code Jam and Kick Start Programming Contest Dataset for Source Code Authorship Analysis (2008–2023)

This dataset contains a structured and anonymized collection of source code submissions from the Google Code Jam and Google Kick Start programming competitions.

The original data originates from publicly available archives hosted at:
https://zibada.guru/gcj/

These archives contain scraped contest data including submissions, participant identifiers, and metadata stored in SQLite databases and compressed archives.

The goal of this dataset is to provide a clean and reproducible corpus suitable for empirical research on:

– source code authorship identification
– programmer stylometry
– machine learning on source code
– behavioral and stylistic analysis of competitive programming solutions

The original archives contain heterogeneous structures and database formats that are not directly usable for large-scale analysis. For this reason, an automated processing pipeline was developed to extract, normalize, anonymize, and reorganize the submissions into a consistent filesystem structure.

The processing pipeline performs the following steps:

1. Extraction of source code submissions from Google Code Jam and Kick Start archives
2. Conversion of SQLite-based archives into a file-based dataset structure
3. Organization of solutions by participant and task
4. Normalization of task names and programming language file extensions
5. Anonymization of participant identifiers
6. Generation of dataset statistics

The resulting dataset structure is organized as follows:

participant_folder/
    task_folder/
        submission_files

Participant folders originally follow the format:

<task_count>-<user_or_id>

After anonymization they are renamed to:

X-id-N

Task folders contain normalized task identifiers derived from contest problem names.

Submission files contain attempt metadata and normalized file extensions (for example Python variants are unified to “.py”).

Each archive corresponds to a single contest year and is distributed as a separate ZIP file.

Included archives:

gcj-2008-2017-dataset.zip  
gcj-2018-dataset.zip  
gcj-2019-dataset.zip  
gcj-2020-dataset.zip  
gcj-2021-dataset.zip  
gcj-2022-dataset.zip  
gcj-2023-dataset.zip  
kickstart-2019-dataset.zip  
kickstart-2020-dataset.zip  
kickstart-2021-dataset.zip  
kickstart-2022-dataset.zip  

Important limitation:

Some older Google Code Jam archives (2008–2017) contain only scoreboard metadata and do not include source code files. In such cases the extraction process can only recover data if source code blobs are present in the archive database.

The dataset is intended for research and educational use.