Citation Audit
Automated two-pass verification of every external reference across 47 published works. 285 unique citations extracted, cross-referenced against CrossRef, doi.org, and Open Library. Every reference is real. Every source exists.
SECS Sentinel Verified denotes verification by the internal audit pipeline that preceded citation-sentinel, the open-source tool built from this methodology. The methodology is identical; the internal version was purpose-built for the SECS corpus.
Validation Pipeline
## References section across 44 papers. Extracted author, year, title, journal, DOI, URL from each entry. Canonical deduplication by author surname + year.doi.org/{doi} for every DOI. Accepted 2xx/3xx as valid. Classified 403/406 responses as paywall-confirmed (DOI exists, content behind access control).Deep Verification Verdicts
50 references that lacked a DOI were subjected to multi-strategy deep search. Each reference was independently located, its role in the citing paper classified, and a falsifiability verdict assigned.
The single UNVERIFIED entry (kreimer_2000) is a parser artefact:
the author and title were embedded in a compound citation string.
CrossRef correctly matched the paper (Connes & Kreimer, 2000,
Renormalization in Quantum Field Theory and the Riemann-Hilbert Problem)
but the score fell below threshold due to the malformed title field.
Citation Roles
Each deep-verified reference was classified by its role in the citing paper:
Most-Cited References
| Reference | Papers | Authors | Year | DOI |
|---|---|---|---|---|
| fan_2023 | 12 | Fan, X., et al. | 2023 | 10.1103/PhysRevLett.130.071801 |
| morel_2020 | 11 | Morel, L., et al. | 2020 | 10.1038/s41586-020-2964-7 |
| parker_2018 | 9 | Parker, R. H., et al. | 2018 | 10.1126/science.aap7706 |
| banach_1922 | 8 | Banach, S. | 1922 | 10.4064/fm-3-1-133-181 |
| tiesinga_2025 | 6 | Tiesinga, E., Mohr, P. J., et al. | 2025 | 10.1103/RevModPhys.97.025002 |
| duncan_2016 | 5 | Duncan, F.E. et al. | 2016 | 10.1038/srep24737 |
| hadley_1997 | 5 | Hadley, M.J. | 1997 | 10.1088/0264-9381/17/20/303 |
| lee_1956 | 5 | Lee, T.D. & Yang, C.N. | 1956 | 10.1103/physrev.104.254 |
| mattauch_1934 | 5 | Mattauch, J. | 1934 | 10.1007/bf01342557 |
| wu_1957 | 5 | Wu, C.S. et al. | 1957 | 10.1016/b978-0-08-006509-0.50011-9 |
Methodology
Primary Audit
- PARSE Regex extraction of structured reference entries from Markdown papers
- CROSSREF REST API queries to
api.crossref.org/workswith author + title + year matching - HTTP DOI resolution via
doi.orgGET requests with browser user-agent - 403/406 responses classified as paywall-confirmed (DOI exists, content behind access control)
Secondary Deep Verification
- CROSSREF Four-strategy search: author+title+year, title-only, author+journal, author+year-broad
- OPENLIBRARY Book lookup via Open Library Search API for textbook citations
- CONTEXT Citation context extraction — surrounding text analysed for how each reference is used
- CLASSIFY Role classification: Foundational, Data Source, Methodological, Contextual, Narrative, Secondary
- VERDICT Falsifiability assessment: Verified, Likely Real, Unverified, Suspicious
Reproducibility
Both audit scripts are deterministic. No random sampling. No manual curation. The full pipeline can be re-run at any time against the live corpus:
python __citation_audit.py --full— primary extraction, lookup, validation, reportpython __citation_secondary_audit.py— deep verification of references without DOI
Source code:
github.com/JustNothingJay/SECS_Sovereign
Open-source tool (generalised from this methodology):
github.com/JustNothingJay/citation-sentinel