slr-automation: code, notebooks and exercises for automating systematic literature reviews

First public release of the code, notebooks and exercises that accompany the chapter on automating systematic literature reviews. Full-text eligibility screening lives in the companion repository FULL_TEXT_SCREENING, pinned here at v0.1.0. What is included Colab notebooks (examples/chapter/notebooks/): one per chapter exercise, S4 to S12 (query expansion, multi-database search, deduplication, screening, full-text retrieval, quality appraisal, data extraction, synthesis and topic modelling, reporting). Exercises and reference solutions (examples/chapter/): S3 to S12, with synthetic fixtures and offline tests, so everything runs without API keys or cost. slr_pipeline/: the LangGraph pipeline behind the chapter's worked example (search backends, screening, full-text, extraction, synthesis, PRISMA-style reporting, cost caps, replay). Thabit2025 worked example (test_reviews/thabit2025/): the lean reproduction used as the chapter's practical example. It contains metadata and short quotes only, no full texts. Offline test suites for both the exercises and the pipeline. Getting started Colab: open a notebook from examples/chapter/notebooks/. Local: git clone https://github.com/rimraul/slr-automation.git cd slr-automation pip install -r requirements.txt pip install -e ".[topics,pdf,viz,dev]" # optional extras pytest examples/chapter -q pytest tests -q Run the two test suites in separate pytest sessions. Requires Python 3.12. AGENTS.md describes how to adapt the code locally, including with a coding agent. Notes Nothing calls a paid API by default. Live calls need your own keys in environment variables (Colab: Secrets). Never commit keys. Datasets other than Thabit2025 that were used during development are not included, so a few development-only tests are excluded from this repository. Licences: code MIT, fixtures and exercise text CC BY 4.0. PyMuPDF (a FULL_TEXT_SCREENING dependency) is AGPL-3.0. Provided as-is. Issues and pull requests may go unanswered; forking is encouraged (CONTRIBUTING.md). Citation See CITATION.cff. A DOI will be attached by the Zenodo archive of this release.