SkillMedev Citation Tracker
Ensure every citation is accurate, complete, and correctly formatted for the requested style. Accuracy comes before formatting - a beautifully formatted wrong citation is still wrong, and the costly failure this skill prevents is the fabricated or unverifiable reference that survives to publication and detonates the author's credibility on contact with one curious reader.
Operating procedure
Verification precedes formatting, because formatting effort spent on a citation that turns out to be retracted or nonexistent is wasted twice.
Step 1: Gather inputs
Collect: the reference list and the manuscript (both - cross-matching needs both sides), the target style, and the venue's house rules if any. If the style is unstated, ask: APA 7, MLA 9, Chicago (notes-bibliography or author-date), or IEEE. Never mix styles in one document, and for Chicago, pin down which variant before touching anything.
Step 2: Inventory the required elements
For each source identify: author(s), year/date, title, container (journal/book/site), volume/issue, pages, publisher, DOI/URL, and access date where the style requires it. A missing element is a flag to report, never something to invent or approximate.
Step 3: Verify before formatting
- Check that each DOI resolves and lands on the cited title - a resolving DOI attached to the wrong paper is the most common fabrication signature.
- Confirm author names and publication year against the source itself, not against another paper's citation of it (citation errors propagate through copying).
- Check retraction status; a retracted source can only be cited as retracted, with that status stated in the text.
- Flag suspected predatory-journal sources - no credible indexing, hijacked titles, implausible turnaround - for the author's judgment.
- Cross-match: every in-text citation has a reference-list entry and every entry is cited in the text. Orphans on either side are findings.
Step 4: Tier the sources by evidence quality
Rate each source so the author sees what their argument stands on:
- Tier 1 - primary, peer-reviewed: original studies and primary data in peer-reviewed venues. The strongest support for an empirical claim.
- Tier 2 - peer-reviewed synthesis: systematic reviews, meta-analyses, scholarly reviews. Excellent for the state of a field; secondary for any specific result (cite the original where a single finding carries weight).
- Tier 3 - preprints and grey literature: not yet peer-reviewed. Citable with an explicit caveat in the text ("a preprint reports...") and never as sole support for a load-bearing claim.
- Tier 4 - non-scholarly: news, blogs, corporate pages. Fine for events and context; not evidence for scientific claims.
Flag any load-bearing claim resting entirely on Tier 3-4 sources.
Step 5: Apply style rules precisely
Key differences to get right:
- APA 7: author-date in text (Smith, 2020); alphabetical reference list; sentence case for article titles; DOI presented as a URL.
- MLA 9: author-page in text (Smith 23); Works Cited; title case; the "containers" model.
- Chicago: notes-bibliography uses footnotes plus bibliography; author-date uses (Smith 2020) like APA. One variant per document, consistently.
- IEEE: bracketed numbers [1] in order of first appearance; reference list numbered, not alphabetical; abbreviated author names.
Then the fine grain: italics vs. quotation marks, comma vs. period separators, "and" vs. "&" rules, hanging indents, and et al. thresholds all differ by style - apply the exact rule, not a plausible average of styles.
Contrast pair (APA 7):
Bad: Smith, John. "Effects Of Sleep On Memory." Journal of Cognition, 2020, https://doi.org/10.1234/fake - full first name (APA uses initials), title case on the article, missing volume/pages, unverified DOI.
Good: Smith, J. (2020). Effects of sleep on memory consolidation in adults. Journal of Cognition, 12(3), 145-162. https://doi.org/10.1234/jcog.2020.0312 - initials, sentence case, italicized container, volume(issue) and pages present, DOI verified to resolve to this title.
Deliverable
Produce: the corrected reference list in the chosen style, and an issues report listing broken or mismatched DOIs, missing elements, retracted or predatory-journal flags, orphaned in-text citations or reference entries, the evidence-tier table, and every source that could not be verified - marked for the author's review, not silently repaired or removed.
Do NOT
- Do NOT fabricate a citation, DOI, page number, or any missing element - flag the gap instead; an invented "plausible" DOI is worse than an honest blank.
- Do NOT mix citation styles, or Chicago variants, within one document.
- Do NOT conflate "formatted correctly" with "verified as real" - report both statuses separately for every entry.
- Do NOT silently drop entries that fail verification; preserve the author's source list and mark failures for review.
- Do NOT trust a citation because another paper cites it the same way - verify against the source itself.
Quality bar
- Every entry carries two explicit statuses: verified (yes / no / partial) and formatted (pass / corrected).
- Every DOI in the final list resolves to the cited work.
- Zero orphans: in-text citations and reference entries match one-to-one.
- Every source has an evidence tier, and Tier 3 sources carry a preprint caveat where cited.
- The issues report exists even when empty ("all N entries verified") so the author knows checking happened.
Escalation
Verifying that a citation is real is not verifying that the claim it supports is true - route claim-level checking to fact-checker or claims-verifier. For weighing conflicting evidence across sources, route to literature-review or systematic-review.