Quotes are verified server-side
No language model tells us where a quote sits. Character positions are looked up in the extracted manuscript text. A quote that does not occur there is shown in the report as unverified.
Research project
Reporting checklists for veterinary manuscripts, with the evidence quoted from your text.
VetCaseAudit is a research project. Veterinary manuscripts are checked against reporting checklists. Every finding is tied to a quote from the submitted text. Quotes that cannot be located in the text are shown as unverified. They are not treated as found.
The running instance is used internally. This is not a commercial offering.
In four steps
From manuscript to evidenced report
- Upload A PDF with a text layer, or a DOCX. Scanned PDFs without a text layer are refused with a specific error message instead of producing an empty analysis.
- Choose a checklist Every analysis is pinned to a published, immutable checklist version.
- Run the screen The text is split into overlapping chunks, each chunk is screened against every checklist item, and the partial results are merged deterministically.
- Read the report One verdict per item, with its supporting quote, a page reference and a concrete improvement suggestion. Exports as PDF and as JSON.
- Presentcounts 1
- Partialcounts 0.5
- Missingcounts 0
- Not applicabledrops out of the denominator
Why the findings hold up
Three rules that are not negotiable
How the screening works
A failed chunk fails visibly
If one chunk of the analysis fails, the whole analysis fails. No item is scored as “missing” in the background. That would push the score down without anyone noticing.
Published checklist versions are immutable
Every analysis stays pinned to the version it ran against. A later update to the checklist does not change a report that already exists.
The checklists
Four checklists: versioned data, not hard-wired logic
STROBE-Vet
Cohort, case-control and cross-sectional studies.
Vet-CASE Core
Developed for VetCaseAudit.
NEURO-5D
Built around the “5 Fingers” framework. Item text is our own.
ARRIVE 2.0
Organised into the “Essential 10” and the “Recommended Set”.
Where your manuscripts are processed
Processed at the GWDG, on your institution's own key
Running an analysis sends the manuscript text to a language-model provider. At present that is only the academic infrastructure of the GWDG in Göttingen, Germany. Anyone using the application stores their own access key. The usage relationship with the provider is between the organisation that uses the tool and GWDG.
Limits
What VetCaseAudit does not do
- It does not replace peer review and makes no scientific, methodological or veterinary judgement. It checks whether something is reported, not whether it is correct.
- The verdicts in the report are AI-generated and can be wrong. They are a working basis for authors to check. The supporting quotes exist for that.
- It is not a plagiarism checker, not a statistics reviewer and not a mass screening tool.
- It does not predict acceptance by a journal and promises no publication outcome.
Sign in
If you already have an account, you can sign in to the application. How the screening works is described under Method.