MARVA

MARVA is a pre-submission audit assistant for journal authors. Before a manuscript goes to a journal, MARVA reads it and reports what needs fixing in each dimension separately. There is no single overall score. The problem is ordinary and well documented. Reference errors are common even in careful work: a 2024 study of researchers who themselves study reference accuracy found 21.2% of references still contained errors. Authors usually discover these problems only after submission, when a reviewer points them out or a desk rejection arrives. MARVA checks a manuscript in stages. It extracts claims, theories, and citations; matches in-text citations against the reference list and flags what is missing, uncited, or duplicated; then checks each reference against Crossref and OpenAlex. A reference counts as verified only when a database record matches its DOI. When the evidence is not enough, MARVA says inconclusive instead of guessing. It also looks for gaps: theories the manuscript relies on but never states, outdated references, and claims that outrun the evidence. The output is a prioritised fix list, strengthening references the author can consider, and journal suggestions where the evidence supports them, plus an exportable report in PDF, JSON, or HTML. Nothing is released as final until the author confirms it. MARVA was verified live on 20 test papers. Its citation matching agreed exactly with hand-checked predictions on all 20, and its reports carry complete counts, including the cases it could not resolve.