Paper Scoring Pipeline

Cross-platform composition layer for Paper Scoring. It combines the core, connectors, and reporting distributions without introducing Git or local-path dependencies.

Configuration is injected at process start. The package never automatically loads .env during import.

pip install paper-scoring-pipeline==0.1.0
paper-scoring-pipeline --env-file .env score-json paper.json --output score.json

Score a local PDF:

paper-scoring-pipeline --env-file .env score-pdf paper.pdf --author "Example Researcher"

Run the arXiv workflow without shell-specific date commands:

paper-scoring-pipeline --env-file .env arxiv --category cs.AI --days-ago 2 --max-results 10

The same command works in Ubuntu shells, Windows PowerShell, and the provided Docker image.

All inter-repository dependencies resolve from PyPI. uv.lock records hashes for development and CI, and the container installs its locked runtime set with hash verification before installing the local pipeline wheel.

Configuration is read only from the process environment and an explicitly selected --env-file; importing the package never reads .env. Errors in a provider, source, report, or SMTP operation are returned as bounded structured failure metadata. Exception messages and tracebacks are not written to the JSON output.

The image runs as UID/GID 65532. Compose also drops all capabilities, enables a read-only root filesystem, and prevents privilege escalation. To verify an offline command with networking disabled:

docker run --rm --network none --read-only --cap-drop ALL \
  paper-scoring-pipeline:0.1.0 --help

The pipeline needs network access for arXiv, OpenAI, OpenAlex, Ollama, or SMTP. For an entirely offline invocation, set PAPER_SCORING_NETWORK_MODE=none when using Docker Compose.