Overview
By the time a thesis, article or policy report is finished, its numbers have passed through several hands and versions: exported from software, pasted into tables, rounded in the text, summarized in the abstract. Its claims have been paraphrased from sources, sometimes from secondary summaries rather than the originals. Each step is a chance for small discrepancies that reviewers, examiners or journalists may notice.
Our verification works in two directions. For claims, we locate the cited source and check that it says what your sentence says it says — the figure, the population, the year, the strength of the conclusion. For data, we check internal consistency: that sample sizes add up, percentages match counts, test statistics agree with degrees of freedom and p-values, and the same number is reported identically in the abstract, results and tables.
Where you can share your dataset and analysis files, we can go further and reproduce key figures. Where you cannot, we still catch a large share of errors by checking the document against itself and against its sources. You receive an itemized log; corrections and any decisions about interpretation remain yours.
Who it’s for and the problems it solves
This is for you if you are…
- Authors preparing a manuscript, especially quantitative papers with many tables
- Doctoral candidates doing a final consistency check before thesis submission
- Research teams, NGOs and think tanks publishing policy reports and briefs that cite statistics
- Grant applicants whose case for support relies on prevalence figures or market data
- Editors and supervisors who want an independent check on a document they are responsible for
Common problems we help with
- Numbers in the abstract, results text and tables do not quite agree
- You cite statistics from reports that may have been updated since you first wrote the section
- A reviewer said a claim overstates what the cited study found
- You are unsure whether reported p-values match the test statistics and degrees of freedom
- The document was written by several people and nobody has checked it end to end
What’s included — and what isn’t
Included
- Claim-to-source checking for all or an agreed sample of factual claims: does the source exist, is it the original, and does it support the specific statement
- Currency checks on official statistics (for example WHO, World Bank, OECD, national statistics offices) with notes where newer releases exist
- Internal consistency of numbers across abstract, text, tables, figures and appendices
- Arithmetic checks: totals, percentages, participant flow, rounding
- Statistical reporting consistency: test statistics, degrees of freedom and p-values recomputed where the reported values allow it
- Checks that figures (axes, labels, values) match the tables and text they illustrate
- Optional reproduction of key results from your data and code
- Notes on overstated, understated or causally worded claims that go beyond what the evidence shows
Not included
- Altering, generating or 'correcting' data to produce different results
- Re-running your entire analysis or replacing your methods (available through our data analysis services)
- Endorsing or certifying the document's conclusions
- Legal review of defamation or regulatory claims
What you receive
- Verification log
- Every checked claim or number with its location, what we checked it against, the result (confirmed, discrepancy, unsupported, outdated, unverifiable) and a suggested fix.
- Consistency report
- A summary of mismatches between sections, tables and figures, grouped so recurring causes (such as a table from an older model run) are easy to spot.
- Statistical reporting check
- A table of reported test results with recomputed p-values where possible and flags where the numbers are inconsistent.
- Reproduction note (optional)
- If you share data and code, a note confirming which key figures reproduce, and details of any that do not.
How it works
Agree the scope
We agree what is checked: all claims, a risk-based sample (headline figures, abstract, executive summary) or only statistical reporting.
Source and data collection
You share the document, cited sources where you have them, and data or output files if reproduction is in scope.
Verification
We trace claims to sources, cross-check numbers throughout the document and recompute what can be recomputed.
Log and review
You receive the log and consistency report, then a call to discuss discrepancies that need a judgment from you.
How we check claims and numbers
- Claim tracing
- Each claim is traced to the cited source and, where that source is itself citing something, to the original. We check population, setting, year, measure and wording of the conclusion.
- Official statistics currency
- Figures from agencies are compared against the current release, and we note when a series has been revised or superseded.
- Statistical consistency checks
- Recomputing p-values from reported t, F, chi-square or z statistics and degrees of freedom, the principle behind tools such as statcheck, and checking that means and SDs are plausible for the scale and sample.
- Cross-document reconciliation
- Building a list of every key number and checking each occurrence, including abstract, highlights, tables, figures and supplementary files.
- Participant flow checks
- Confirming that recruitment, exclusion and analysis numbers reconcile, as in a CONSORT or PRISMA flow diagram.
- Reproduction
- Running your code on your data in R, Stata, SPSS or Python to confirm the reported figures, without changing the analysis.
When a claim drifts from its source
Citation drift is common and usually innocent. A study of adolescents becomes evidence about 'young people'; an association becomes an effect; a figure from one country is presented as global; a 2015 statistic is described as current. Each is a small overstatement that a careful reviewer will question.
We flag these with the exact wording from the source so you can decide whether to narrow the claim, update the figure or find stronger evidence. The judgment and the rewrite stay with you.
Numbers that disagree with each other
Discrepancies between the abstract and the results, or between a table and its description, often come from rerunning an analysis after the text was written. Our reconciliation lists every occurrence of each key number, making it obvious which version is out of date.
Statistical reporting gets its own pass. If a paper reports t(48) = 2.10, p = .002, the numbers cannot all be right; the p-value for that statistic is about .04. We recompute what can be recomputed and flag the rest for you to check against your output.
- Sample sizes that change between sections without explanation
- Percentages that do not match counts, or subgroups that do not sum to the total
- Confidence intervals that do not contain the point estimate
- Figures whose plotted values differ from the table
What the work can look like
Illustrative example — fictional data
Sample excerpt: verification log
A research team asked for a check of a 40-page policy report on youth unemployment before publication. The entries below are invented to show the log format.
Claim, page 3
'Youth unemployment has doubled since 2015.' The cited national statistics release shows an increase from 14.2% to 21.9%, about 54%. Suggested fix: report the actual figures or 'risen by about half'.
Number, pages 1, 12 and Table 4
Sample size appears as 1,204 in the executive summary, 1,240 in the methods and 1,204 in Table 4. Please confirm the correct figure against your data file.
Statistic, page 18
Reported chi-square(2) = 3.1, p = .01. For these values p is about .21. Please check the output; either the statistic or the p-value appears to have been transcribed incorrectly.
Illustrative example. The report, figures and discrepancies are invented and do not describe real client work.
What you’ll need to provide
- The document in an editable format, with tables and figures
- PDFs or links for cited sources, especially paywalled or grey literature
- Statistical output (SPSS, Stata, R, Excel) behind the reported results, if available
- Data and code if you want key figures reproduced
- Which claims matter most to you, if a sample-based check is preferred
You don’t need everything ready to ask for a quote — tell us what you have and we’ll say what’s missing.
What affects price and turnaround
Price depends on
- Document length and the number of claims, tables and figures
- Whether checking covers every claim or an agreed sample
- Accessibility of sources (paywalled, archival, non-English, grey literature)
- Whether reproduction from data and code is included
- Deadline
Turnaround depends on
- Availability of source PDFs and statistical output at the start
- Scope: full check versus sample of headline claims
- Time needed for you to answer queries about discrepancies
Every project is quoted individually. How we price our work.
Questions people ask before starting
Do I have to share my data?
No. Most checks work from the document and its sources. Sharing data and code only matters if you want key figures reproduced, and it can be anonymized first.
What happens if you find a serious error?
We tell you promptly, with the evidence, so you can decide how to correct it. For published work, we can explain how corrections are usually handled, but the decision to contact a journal is yours.
Can you check every sentence of a long thesis?
We can, but a risk-based sample (abstract, key findings, headline statistics, contested claims) often gives most of the value in less time. We will suggest a scope based on your document and deadline.
Is this the same as a citation audit?
They overlap but differ. A citation audit verifies that references exist and match; fact verification checks whether sources support the specific claims and whether your numbers are consistent. Many clients combine both.
Will you keep my unpublished data confidential?
Yes. Documents and data are used only for the check, stored securely for the duration of the work and deleted on request.