Living Review Documentation

This project provides tools for building living reviews of machine learning and accelerator physics publications. It includes:

• Fetchers for arXiv, InspireHEP, HAL, OpenAlex, Crossref.

• A staged relevance funnel: metadata enrichment, deterministic gates, and a zero-shot NLI adjudicator with a human pending queue.

• Identifier-graph + fuzzy-title deduplication.

• Category classification and statistics computation.

• Export to JSON (published site data), BibTeX, and PDF.

• A full pipeline + CLI for automation.

Contents

Methodology

• Methodology: the relevance funnel
  • Overview
  • Stage A — Fetch, enrich, deduplicate
  • Stage B — Deterministic gates
  • Stage C — Zero-shot NLI adjudication
  • Terminal decisions and provenance
  • Human curation
  • Quality gates in CI

API Reference

• living_review package
  • Submodules
  • living_review.classifier module
  • living_review.gates module
  • living_review.adjudicator module
  • living_review.relevance module
  • living_review.dedup module
  • living_review.enrich module
  • living_review.migrate module
  • living_review.cli module
  • living_review.config module
  • living_review.data_model module
  • living_review.exporters module
  • living_review.fetchers module
  • living_review.logs module
  • living_review.pipeline module
  • living_review.stats module
  • living_review.utils module
  • Module contents