Citation Verifier — Existence-Verify a Bibliography Against Public Scholarly APIs

The headline existence-verification skill: given a bibliography in any common form, it proves entry-by-entry whether each reference actually exists by querying four keyless public scholarly APIs, then maps each result to the canonical AlterLab citation-hallucination taxonomy. It is the deterministic, network-grounded companion to the LLM-driven integrity_verification_agent — where that agent uses WebSearch + judgment, this skill uses authoritative API records and a reproducible Python script, so the same input always yields the same verdicts.

Quick Start

Verify the citations in references.bib
Check whether these DOIs resolve to the papers I cited
Audit my bibliography for fabricated / hallucinated references
Does this reference list contain any fake citations or retractions?

→ Run scripts/verify_citations.py over the bibliography, read the JSON, then present a verdict table grouped by severity. Always state the offline/degraded status explicitly if the network was unavailable.

WHAT This Does

For each bibliography entry the script:

1. Parses the input (auto-detects BibTeX / DOI-list / free-form), extracting

title, authors, year, venue, DOI, and arXiv ID.

1. Resolves identifiers — looks up the cited DOI/arXiv ID directly when present.
2. Searches by title as a fallback across all four sources.
3. Fuzzy-matches the cited title (difflib SequenceMatcher ratio, default

threshold 0.70) and computes author-surname overlap.

1. Flags retractions marked in Crossref (update-to: retraction) or OpenAlex

(is_retracted).

1. Emits a verdict per entry mapped to the taxonomy below, plus a repo-level

summary.verdict (PASS / PASS_WITH_CONDITIONS / FAIL / UNVERIFIED).

The four sources (all keyless, polite-pool)

All requests carry a mailto parameter (defaults to [email protected]) to stay in each API's polite pool. No API keys are required or used.

WHEN To Use It (and when not)

This skill answers "does the cited work exist, and does its identifier point to it?" It does not read the cited paper's full text, so it cannot by itself confirm Semantic Hallucination (does the source support the claim?) — that requires claim_verification_protocol. SH is surfaced only as an advisory flag, never asserted from API metadata alone.

Verdict Taxonomy (mirrors the canonical Five-Type Taxonomy)

Identical codes and definitions to alterlab-research-pipeline/agents/integrity_verification_agent.md (GPTZero × NeurIPS 2025; Ansari, 2026). Severity feeds the same SERIOUS / MEDIUM / MINOR scale used in the Integrity Report schema.

A RETRACTED flag is attached (and severity bumped to SERIOUS) whenever Crossref or OpenAlex marks the matched work as retracted, independent of the existence verdict.

Repo-level verdict

• PASS — every entry verified, no SERIOUS/MEDIUM flags.
• PASS_WITH_CONDITIONS — only PAC / MEDIUM issues (wrong metadata, fixable).
• FAIL — any SERIOUS verdict (TF / IH / PH / retraction).
• UNVERIFIED — entries are unverified and no concrete fabrication was found

(e.g. fully offline run). This is not a pass — re-run with network access.

Pipeline (how to run it)

1. Locate or capture the bibliography

Accept any of: a .bib file, a .txt list of DOIs/arXiv IDs, a pasted reference list, or inline text. The script auto-detects the format; override with --format bibtex|doi|freeform if detection is wrong.

2. Run the verifier

uv run python skills/core/alterlab-citation-verifier/scripts/verify_citations.py \
    path/to/references.bib \
    --mailto [email protected] \
    --threshold 0.70 \
    --out citation_report.json

• path/to/references.bib may also be - (stdin) or inline text.
• --threshold tunes the fuzzy title-match ratio (0..1; default 0.70).
• --offline skips the network and emits unverified verdicts deliberately.
• Omit --out to print the JSON report to stdout.

The script auto-selects an HTTP backend: it uses requests if installed, else falls back to the Python stdlib (urllib) — so it runs with zero extra dependencies in a bare uv environment.

3. Read the JSON and report

Parse summary.verdict and the per-entry verdict codes. Present:

1. The headline verdict and counts (verdict_counts, severity_counts).
2. A table of every non-`verified` entry with its code, severity, and detail.
3. For each TF / IH / PH: quote the cited entry and explain the evidence

(e.g. "DOI 10.x resolved to an unrelated paper titled '…'").

1. Any RETRACTED flags, prominently.
2. If verdict == UNVERIFIED: state plainly that nothing was confirmed and relay

the per-entry manual_instructions.

4. Route fixes

• TF / PH → the reference must be removed or replaced; it does not exist.
• IH → the DOI/arXiv ID is wrong; find and substitute the correct identifier.
• PAC → correct the specific metadata field(s) named in detail.
• RETRACTED → flag to the author; cite the retraction notice or drop the source.

Graceful Degradation (no network)

Network failures are never silently swallowed into a pass:

• A DNS/connection failure raises NetworkUnavailable; the entry becomes

unverified with a populated manual_instructions field.

• --offline forces every networked entry to unverified up front (placeholders

are still caught locally as PH).

• The repo-level verdict becomes UNVERIFIED (distinct from PASS) whenever

unverified entries exist without any concrete fabrication finding.

When degraded, instruct the user to re-run with connectivity, and fall back to the LLM-driven integrity_verification_agent (WebSearch) for a manual pass.

Output Shape (excerpt)

{
  "tool": "alterlab-citation-verifier/verify_citations.py",
  "version": "1.0.0",
  "summary": {
    "total": 2,
    "verdict": "FAIL",
    "verdict_counts": {"verified": 1, "TF": 1, "PAC": 0, "IH": 0, "PH": 0, "SH": 0, "unverified": 0},
    "severity_counts": {"SERIOUS": 1, "MEDIUM": 0, "MINOR": 0},
    "citation_integrity_score": 0.5,
    "fabrication_risk_score": 0.5,
    "retracted": 0
  },
  "entries": [
    {"ref_id": "walters2023", "verdict": "verified", "severity": "NONE",
     "title_ratio": 1.0, "author_overlap": 1.0, "matches": [{"source": "crossref"}]},
    {"ref_id": "ghostpaper2021", "verdict": "TF", "severity": "SERIOUS",
     "detail": "Cited DOI/arXiv identifier did not resolve in any source..."}
  ]
}

citation_integrity_score and fabrication_risk_score (both 0..1) align with the Integrity Report schema fields of the same name, so the report can feed alterlab-research-pipeline's integrity gate directly.

Self-Check Before Reporting

• Did the run reach the network? If config.http_backend ran but every entry is

unverified, the network was down — say so; do not imply a pass.

• Are there any RETRACTED flags? Surface them even on otherwise-verified entries.
• Did any entry score IH? Confirm the detail shows an id-resolved mismatch,

not a loose title-search coincidence (the script enforces this distinction).

• Is the headline verdict consistent with the per-entry codes (any SERIOUS → FAIL)?

References

• alterlab-research-pipeline/agents/integrity_verification_agent.md — canonical

Five-Type Taxonomy, compound-deception patterns, and the Lin et al. (2020) mashup case study this skill is built to catch.

• alterlab-research-pipeline/references/claim_verification_protocol.md — Phase E

claim-vs-source verification (the SH check this skill defers to).

• shared/schemas/integrity_report.schema.json — the integrity-report shape whose

citation_integrity_score / fabrication_risk_score this skill mirrors.

• Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the

bibliographic citations generated by ChatGPT. Scientific Reports, 13, 14045. https://doi.org/10.1038/s41598-023-41032-5

• Ansari, S. (2026). Compound Deception in Elite Peer Review. arXiv:2602.05930.