# Third-party attributions

## Trafilatura (Apache-2.0)

Content-root candidate token vocabulary in `src/chromerag/content_root.py` is adapted from
Trafilatura's `BODY_XPATH` selectors in `trafilatura/xpaths.py`.
https://github.com/adbar/trafilatura

Discard-pattern vocabulary in `src/chromerag/rules/trafilatura_inspired.yaml` is adapted from
Trafilatura's XPath discard lists in `trafilatura/xpaths.py` (Apache-2.0, Adrien Barbaresi).

## WCXB (CC-BY-4.0)

The evaluation scripts in `poc/` read the Web Content Extraction Benchmark (M. Foley, arXiv:2605.21097,
https://arxiv.org/abs/2605.21097, licensed CC-BY-4.0). The model in `src/chromerag/assets/` was trained on pages and
human-reviewed labels from its development folder. The benchmark data itself is not redistributed here.
