Paper processing with parsing, reconciliation, and Colab/R2 automation.
The operational source code is in new_code/. Running it requires Ubuntu/WSL,
an authenticated Colab CLI, private R2 configuration, and the original reference PDFs.
Model weights are downloaded according to the manifests; data, results, and credentials
are kept out of Git.