Manuscript verification
How to audit unsupported claims before submission
A reference list can be accurate while the argument remains weak. A claim audit checks whether each important sentence says only what its cited evidence supports.
In this article
Citation checking is often treated as a formatting task. The author confirms that every bracket has a reference, the reference fields are complete, and the numbering is correct. Those checks matter, but they do not answer the harder question: does the cited source support the sentence that points to it?
The IEEE Author Center tells authors to cite references that directly support their work. The International Committee of Medical Journal Editors goes further: authors should be able to attest that a reference supports its associated statement. These standards apply to the relationship between claim and evidence, not only to the accuracy of the bibliography. See the official guidance from the IEEE Author Center and the ICMJE.
A complete reference is not proof that the sentence beside it is supported.
Audit the argument, not only the bibliography
A manuscript can fail at three different layers. A bibliographic error means the record is wrong or incomplete. A placement error means the citation appears too far from the statement it is meant to support. An evidence error means the source is real and correctly placed, but its methods, population, result, or uncertainty do not justify the wording of the claim.
Citation software can catch many record and numbering problems. Evidence errors require reading. A reliable workflow therefore starts with the prose and moves back to the exact passage, table, figure, or result that supports it.
Build a claim inventory before reading every citation
Do not begin by reopening the reference list from top to bottom. Begin with the manuscript. Mark sentences that report a measurement, compare methods, state a cause, generalize to a population, recommend an action, or use absolute terms such as always, prevents, proves, or eliminates.
Split compound sentences into separate claims. A sentence can contain one supported statement and one unsupported conclusion. Give each claim a short ID, record its section, and note the citation that is supposed to support it. This creates a review queue that can be divided among coauthors without duplicating work.
- Quantitative claims: measurements, rates, confidence intervals, and performance changes.
- Comparative claims: one method is faster, safer, more accurate, or more robust than another.
- Causal claims: an intervention, mechanism, or design choice produces an outcome.
- Scope claims: a result applies across systems, populations, operating conditions, or time.
- Novelty claims: a method is the first, only, or best available approach.
Read the evidence at passage level
Search the cited paper for the variables, outcome, or method named in the claim. Then read beyond the matching sentence. Check the surrounding paragraph, the relevant table or figure, the methods used to produce the result, and the limitations that constrain it. An abstract can help locate a result, but it is rarely enough for a final audit.
Record the page, section, table, or figure that carries the support. Copy a short working excerpt only when your access rights permit it, and keep the source link or PDF beside the note. The goal is to let another reviewer reproduce your decision without searching the entire paper again.
Use four verdicts and one required action
A simple status system keeps the audit consistent. Every claim receives a verdict and, unless it is supported as written, a concrete next action.
| Verdict | Meaning | Required action |
|---|---|---|
| Supported | The evidence matches the wording and scope. | Keep the claim and record the passage. |
| Qualified | The source supports a narrower statement. | Revise the wording to match the conditions. |
| Contradicted | The source reports a conflicting result. | Correct the statement or explain the disagreement. |
| Unsupported | No cited evidence establishes the claim. | Add suitable evidence or remove the claim. |
This system separates writing quality from evidentiary status. A fluent sentence can still be unsupported. A cautious sentence can be accurate even when the underlying study is limited.
Look for scope and strength errors
Population drift
The source studies one population, material, grid model, dataset, or operating region, while the manuscript generalizes to a broader class. Restore the missing boundary in the sentence.
Outcome drift
The paper measures a proxy, but the draft names the desired real outcome. For example, a simulation metric does not by itself establish field reliability. Name the measured outcome and avoid substituting an adjacent one.
Strength drift
The source reports an association, trend, or result under selected conditions, while the draft uses causal or universal language. Replace proves, causes, and always with wording that reflects the study design.
Version drift
A preprint, conference paper, dataset, or standard may have a newer version. Verify the persistent identifier, publication status, and correction or retraction record before submission. The ICMJE specifically asks authors to check for retracted references and to identify preprints clearly.
A practical engineering example
Consider a hypothetical power systems manuscript that says: “Grid forming control prevents instability in converter dominated networks.” The cited paper tests two simulated weak grid cases and reports improved damping for selected modes. The citation is relevant, but the claim is broader than the evidence.
A defensible revision would identify the setting, method, and measured outcome: “In the two simulated weak grid cases, the evaluated grid forming controller increased damping for the selected modes.” A second source may justify a broader comparison, but adding references should not be used to hide a wording problem.
The same test works in medical, social science, and computing manuscripts. Match the claim to who or what was studied, what was measured, under which conditions, and with what uncertainty.
Apply the reporting checklist after the claim audit
Claim support and reporting completeness are related but different. Once the prose matches the evidence, use the reporting guideline required for the study design or venue. The EQUATOR Network maintains reporting guidelines for many health research designs. Other fields use venue instructions, standards, reproducibility checklists, and data or code availability requirements.
A checklist confirms that required information appears. The claim audit confirms that the manuscript does not say more than the evidence permits. Use both before the final formatting pass.
Run the final pass in manuscript order
- Resolve every claim still marked qualified, contradicted, or unsupported.
- Confirm that each citation sits beside the statement it supports.
- Verify titles, authors, years, identifiers, and publication status against the original source or a trusted bibliographic database.
- Check that tables, figures, abstracts, and conclusions use the same scope as the results section.
- Ask a coauthor to reproduce a sample of claim decisions from the saved passages.
- Run the venue checklist and export the final manuscript and bibliography.
Rivul can flag claims without visible support, compare a selected claim with a cited passage, and keep citation records beside the manuscript. It can also run preflight checks before Word or LaTeX export. These tools shorten the review path, but the author still decides whether the source is methodologically adequate and whether the final wording is justified.
For a broader source to manuscript method, read research without losing the evidence. To organize several studies before drafting, use the literature review evidence matrix.