top of page

Editor(s) are responsible for evaluating the integrity of submitted work, including AI-assisted work, and for determining whether the evidence is verifiable, traceable, internally consistent, and sufficient to support the conclusions. Editorial oversight will focus on verification of the scientific evidence rather than detection of the tools used to produce or prepare the work. When concerns arise, the level of editorial review will be proportionate to the nature and significance of the issue identified.

The editorial staff may use appropriate human, computational, and artificial-intelligence-assisted methods to evaluate submitted manuscripts. Verification may include:

● Verification of references, DOIs, PMIDs, trial registrations, datasets, and other source identifiers.

● Assessment of whether cited sources actually support the statements and conclusions for which they are cited.

● Evaluation of consistency among the manuscript text, tables, figures, supplementary material, numerical results, and cited evidence.

● Review of statistical and mathematical coherence, including sample sizes, denominators, effect estimates, confidence intervals, P values, and other reported results where applicable.

● Assessment of whether the Methods provide a sufficient and reproducible pathway to the reported Results.

● Screening for inappropriate duplication of text, data, results, figures, or images.

● Examination of figures and images for inappropriate alteration, duplication, or manipulation.

● Verification of ethics approvals, informed-consent requirements, trial or study registration, reporting requirements, and data-source information where applicable.

● Assessment of data and result plausibility, including unexpected inconsistencies or patterns that may warrant clarification.

● Evaluation of whether the principal conclusions are proportionate to and supported by the evidence presented.

The use of automated or artificial-intelligence-assisted verification does not replace editorial judgment. Material concerns identified computationally should be evaluated in context by an Editor or other appropriately qualified individual.

​

A Tiered Editorial Verification Architecture

Level 1: Universal Verification

Routine checks may be applied to all manuscripts to identify errors, inconsistencies, unsupported claims, citation problems, reporting deficiencies, duplication, and related concerns.

​

Level 2: Trigger-Based Enhanced Review

When routine checks, peer review, or editorial assessment raises a question about reliability, provenance, or reproducibility, enhanced review may include claim-to-citation verification, statistical or methodological assessment, recalculation of results, examination of images or related publications, and requests for supporting documentation. Enhanced review is intended to resolve a scientific or editorial question and does not constitute an allegation of misconduct.

​

Level 3: Source-Data and Provenance Audit

When a material concern remains unresolved, Editor(s) may request relevant source data, analytical code, statistical outputs, original images, protocols, ethics documentation, registrations, dataset provenance, or other records necessary to verify the work. Failure to provide information reasonably necessary to verify a material scientific claim may affect editorial consideration.

 

AI Detection Is Not Scientific Verification

Identification of AI-generated or AI-assisted material does not, by itself, establish scientific reliability or misconduct. AI-detection tools should not be the sole basis for an editorial decision. When concerns arise, Editor(s) should evaluate the underlying evidence, provenance, methods, citations, analyses, and internal consistency of the work.

©2020年,全球新生兒學會,“每個嬰兒都很重要”

bottom of page