Limitations
-
GROBID sub-citations: GROBID misclassifies a small fraction of references, particularly sub-citations with letter suffixes like
Liu et al., 2024b. These are silently dropped. -
Ambiguous narrative citations: The supplementation pass skips ambiguous
(surname, year)signatures (e.g. two different Zhou 2022 papers in the bibliography). Rare but possible in survey papers. -
Sentence splitting: pysbd handles most academic abbreviations but can occasionally split mid-sentence. Falling back to
--context-window 300sometimes helps. -
arXiv rate limits: arXiv enforces ~3 seconds between requests. The first run on a deep trace can take several minutes. Subsequent runs are fast thanks to the local cache.
-
Unavailable papers: Papers not on arXiv, OpenReview, Sci-Hub, S2 open-access, or any supported preprint server appear as red
unavailablenodes. Books and some workshop proceedings are typically not retrievable. Use--supply-pdfto provide PDFs manually (local path or URL). -
Fruchterman-Reingold layout: Implemented via vis.js's
forceAtlas2Basedsolver, which is the closest available approximation. A proper Kamada-Kawai implementation isn't offered because vis.js doesn't ship one. -
Semantic matching model quality: The default
all-mpnet-base-v2was benchmarked at F1=0.93 on academic citation text. Domain-specific keywords may benefit from threshold tuning. Not available in reverse trace mode. -
Diff paper_id instability:
paper_idis not fully stable across runs. A paper resolved by title hash in one run and by DOI in another may falsely appear as "new" in a diff. Using the same cache directory minimizes this.