verifyR.med
verifyR.med checks whether the effect sizes, confidence intervals and tables reported in a medical or medical-education manuscript agree with each other, so a reviewer knows where to look first. Free to use in your browser; your file never leaves your device.
What verifyR.med checks
verifyR.med reads the numbers a paper reports and tests whether they agree with each other: confidence intervals against p-values, impossible values, percentages and table arithmetic, and heterogeneity statistics in meta-analyses. It reads Word, PDF, HTML and JATS files, including tables, and lists what was checked and found consistent, not only what looks wrong.
Effect sizes with confidence intervals
OR, HR, RR, mean difference, Cohen’s d, ICC, kappa, AUC, NNT, sensitivity and specificity are read together with their intervals.
An audit trail
Every scan lists what was read, which checks ran and what was found consistent, so a clean line is never a silent one.
Checks
- Confidence interval against p-value: does the interval agree with the reported p-value for odds, hazard and risk ratios, mean differences, d and other effect sizes?
- Impossible values: an ICC or AUC above 1, a percentage above 100, a negative count
- Percentages against the counts and group sizes they refer to
- Table arithmetic: rows, columns and totals that should add up
- Meta-analysis heterogeneity: I² and tau² against the other reported figures
- Tables in Word, PDF, HTML and JATS files are read and checked as well
- Methods and reporting hints: scale totals against the number of items and response options stated in the Methods, standard deviations against the possible range, and an instrument given different item counts
- Reporting consistency: confidence intervals with a bound printed as zero next to a significance claim, ‘marginal’ wording for p above .10, fit cut-offs attributed to Hu and Bentler (1999), and fit wording against the reported indices
- Unfinished template text, eligibility criteria against the reported age range, group counts against N, reliability not reported for the sample, and causal wording in a cross-sectional design
How to read the results
The scanner sorts what it finds into three tiers, so the strongest signals are not buried among the weaker ones.
FLAG
Strongest signals: a result that contradicts itself or cannot be right as printed, for example an ICC above 1, or a confidence interval that excludes the null value next to a p-value that does not.
REVIEW
Places a person should look; many will turn out fine. For example a p-value that does not fit its interval, percentages or table totals that do not add up, or heterogeneity values that do not match.
INFO
Notes about how something was read: reporting-style remarks and values that could not be parsed.
Every flag is a prompt for human review, not a verdict; some will be false alarms
A flag can be a typo, a rounding artefact, a different model or test, or a reading error. A clean result only means that the numbers the software recognised were consistent. Decisions about a manuscript belong to people who can see the data.
Your file stays on your device
- The scanner runs on your device. The manuscript is read in your browser and is never uploaded; the site only serves the page and its static files.
- You can verify this: the scanner has a "Prove it" panel that lists every request the page made, and your browser developer tools show the same.
- The micro-tools also run in your browser. One exception: the DOI checker sends the DOIs you paste (not the reference text) to the public Crossref API, and says so on its page. Do not use it for confidential reference lists.
- Local processing is a design property, not a guarantee against malware, browser extensions or a compromised host. For highly sensitive manuscripts, check your own setup.
In-browser manuscript scanner
Drop a .docx, .pdf, .html, JATS .xml or .txt file; tables are read too. Results appear in the FLAG, REVIEW and INFO tiers, and every scan lists what was read, which checks ran, and what was checked and found consistent. The statistics engine runs as R compiled to WebAssembly inside your tab. A scan that finds too few checkable statistics is shown as “nothing checkable found”, never as clean. The first visit downloads the analysis runtime; later visits start from the cache.
Open the scannerChecking a psychology manuscript with APA test results? See verifyR
Free micro-tools
Four small tools that run client-side. Each flag is a prompt for human review.
CI and p consistency
Does the confidence interval agree with the p-value? Operator-aware, with an approximate implied p.
Open toolGRIM test
Can a reported mean arise from whole-number data with this N? Shows the nearest attainable means.
Open toolAPA p-value format
Finds p-values in text and fixes .000, leading zeros, spacing and scientific notation.
Open toolDOI and reference check
Does each DOI exist in Crossref, and is it registered under the title you cite?
Open toolScope
- No OCR in the browser: scanned or image-only PDFs yield no text.
- It reads effect sizes and intervals written in common reporting form. A value described only in words, or one split from its interval by other text, is outside its scope.
- Tables in PDFs are recovered from the page layout on a best-effort basis. Word, HTML and JATS tables are read from their structure.
- Figures such as forest plots, images and equations stored as images are not read.
- APA-style test results of psychology manuscripts (t, F, r, chi-square, z) are handled by verifyR.
- It supports review; it does not replace peer review, statistical review or reading the paper.
- Study design, measurement validity, model specification and whether cited sources support the claims are not assessed. A clean result is not a statement that the paper is sound.
Citation and contact
Suggested citation
Kara, E. verifyR.med: Statistical verification for medical and medical-education manuscripts [Computer software]. University of Aberdeen.