This page turns CIDM iReview into a practical pre-submission resource: clearer denominators, safer causal language, better validation claims and more consistent tables. The review is living and correction-aware. It is not a scorecard of authors, teams or papers.
The most useful output of a large literature review is not a list of faults. It is a reusable set of checks that helps future papers become clearer, easier to review and less likely to need correction later.
Current effective signal prevalence among substantively screened papers: 98 / 963 = 10.2%. This is a review-signal rate, not an error rate.
Reviewer caution ≠ manuscript errorStudy limitation ≠ writing errorNot visible to us ≠ missing from the paperNeeds validation ≠ invalidCitation ≠ evidence
Totals, subgroups, percentages and analysed N should agree.
Patients, samples, isolates, tests and episodes are not interchangeable.
Retrospective association or model fit is not prospective clinical utility.
Temporal sequence, observational association or ex-vivo function does not by itself establish a causal clinical effect.
Negative detection does not prove zero prevalence or zero risk.
Sequence similarity can support linkage hypotheses but does not prove transmission direction on its own.
Keep genotype-predicted and phenotype-confirmed resistance separate, with the confirmation denominator visible.
Agreement with another assay is not independent validation against a reference standard.
Keep tested, available and confirmed subsets attached to their exact numerator and denominator.
Abstract, main text, tables, figures, supplements and specimen IDs should tell the same numerical story.
The raw screen is historical evidence about the review process. The correction ledger is part of the scientific record, not an embarrassment to hide.
The public page deliberately keeps teaching examples anonymised. Named attribution should wait for stronger source return, author clarification, or a clear scientific reason that identification improves understanding. The default use is education, manuscript quality improvement and transparent correction — not fault assignment.
AI review remains preliminary scientific assistance. Human Gate is OPEN; Formal Accept = 0. Author confirmation is representation evidence, not a truth verdict.
Corpus: 1,947 exact publication objects. First-pass Deep iReview: 1,947 / 1,947 closed. Active depth: 258 FULL · 1,544 ABSTRACT · 145 METADATA. The writing-issue living audit currently uses 963 substantive screens and 984 metadata/title triage records; these are different layers and should not be conflated.
Public statistics on this page inherit the living audit and reverse-audit ledger. They should be updated when the underlying evidence changes; do not independently recompute them for a slide or manuscript.