EpiLens: Lightweight RAG for Epidemiological Knowledge Acquisition. Results, code, and corpus references

Reproducibility package for EpiLens: frozen results, analysis code, and experiment code. Contents recorded/: frozen per-paper predictions (994 run files), gold labels for the 214-paper evaluation set, and the recorded metrics and statistical tests. It also holds the paper's tables and figures, the list of the 2,296 classified papers with bibliographic metadata, and the review-expansion bibliography database. src/: code that regenerates and checks the reported results, implements the metrics and baselines, and documents the commands used for every reported run. Reproduce the reported results (Python 3.10; no network, APIs, databases, or GPU needed after installation): python -m venv .venv source .venv/bin/activate python -m pip install -r requirements-lock.txt make check make check recomputes the metrics and regenerates the tables and figures, then compares metrics to 1e-12 and tables exactly against the recorded copies. README.md maps each paper table and figure to its file and generator. Rerunning the experiments requires the documents loaded into MongoDB and Qdrant, which are not included, plus models or API credentials; see src/rerun/README.md. EpiLens is installed from https://github.com/VinsRR/EpiLens at commit 146510d. Not included: article PDFs, full text or abstracts, document databases, vector indexes, model checkpoints, prompts, retrieved passages, or model traces. Licenses: code and configurations MIT; authored data and derived results CC BY 4.0 (see RIGHTS.md).