The turbulent integration of generative language models into the research cycle has prompted an institutional response, including stricter editorial policies and ubiquitous AI detection tools. This article analyzes the statistical limitations of automated detectors and their effect on academic freedom and cognitive diversity in scientific discourse. Recent empirical data (2023–2026) indicate that the false-positive rate is approximately 16% when testing a single detector using texts written by native English-speaking students. However, this figure rises to 61.3% for texts written by non-native English speakers when testing seven detectors, indicating a systemic bias in these tools. Bayesian statistics suggest that, due to the low prevalence of violations, even a moderate proportion of false positives can lead to a high false discovery rate (FDR) in sanction decisions. Legitimate scenarios for AI application during the stages of literature review, hypothesis formalization, methodological design, and linguistic preparation of the manuscript for publication are considered. Approaches to regulating the use of AI in scientific research and the presentation of research findings in the form of scientific texts have been systematized using official statements from COPE, WAME, Nature Portfolio, American Association for the Advancement of Science (AAAS), Elsevier, Springer Nature, Institute of Electrical and Electronics Engineers (IEEE), Association for Computing Machinery (ACM), and the International Federation of Translators (FIT), as well as a review of relevant publications. The authors argue the need to transfer from a culture of detection and prohibition to that of transparent disclosure, cryptographic traceability, and verification of the research process. Particular attention is given to the Russian context, including the implementation of an AI-detection module in the Antiplagiat system and the absence of a unified stance by the Higher Attestation Commission (VAK) and the White List operators. The applicability of Western quality metrics for plagiarism detectors to Russian-language texts is also discussed. Another focus is the transformation of scholarly text creation and the peer-review system, the global inequality of access to language tools, as well as the “arms race” between detection tools and circumvention technologies.

Due to the rapid growth in the number of scientific publications and increasingly stringent quality requirements, journal editorial offices are in greater need of effective tools to verify reference lists. Existing solutions are either commercially unavailable for regional journals or provide fragmented information without the possibility of comprehensive analysis. This paper presents an open-source software tool, Comprehensive Reference List Analysis (CRLA), available at https://litanalysis.streamlit. app and developed in Python using the Streamlit framework. The aim of this article is to demonstrate the functional capabilities of CRLA as a ready-to-use solution for automated reference list analysis and to illustrate its practical applicability using several articles as examples. The tool integrates the Crossref and OpenAlex APIs and enables automatic identifier extraction, metadata enrichment, duplicate detection, suspicious DOI identification, retracted article detection, repository source identification, as well as the calculation of bibliometric indicators and visualization of results. To demonstrate the functionality of the tool, the reference lists of three articles (published in the journal Scientific Editor and Publisher in 2025 and 2026) were analyzed. The obtained results show that CRLA can reveal structural features of reference lists depending on article subject matter: ranging from a predominance of web sources in reviews of publishing practices to nearly complete DOI coverage in bibliometric studies. The tool also identifies self-citations, international collaborations, and potentially problematic sources (duplicates, incorrect DOIs). Thus, CRLA offers a convenient solution, especially for journals with limited resources, enabling the automation of routine checks and improving the quality of peer review.

Author keywords are used in the search, indexing and thematic classification of scholarly publications, yet existing studies treat their overlap with the title as an aggregate measure and do not offer a reproducible calculation procedure at the level of individual manuscripts and the corpus. This article examines token-level overlap between author keywords and titles of scholarly manuscripts. The purpose of the study is to develop and apply a reproducible procedure for calculating this overlap at the level of individual manuscripts and the corpus as a whole. The material consisted of 779 English-language manuscripts submitted to an international scholarly journal in 2022–2026. The analysis used manuscript titles and author keywords. Preprocessing included text normalization, tokenization, identification of overlapping and new tokens, calculation of TKOI (Title Keyword Overlap Index), the complementary TKEI (Title Keyword Expansion Index), and absolute token counts. The calculations were performed in Python using the pandas, NLTK, and matplotlib libraries. In the main preprocessing mode, 43.4% of keyword tokens overlapped with title tokens, while 56.6% were absent from titles. With the two alternative preprocessing modes, the share of overlapping tokens ranged from 42.9% to 46.1%. Across all three variants, new tokens outnumbered overlapping tokens. In 34 of 779 manuscripts, surface token overlap with the title was completely absent. The median TKOI was 0.444444, and the interquartile range was 0.314286. The proposed procedure can be used in corpus-based analysis of scholarly metadata to examine the relationship between author keywords and manuscript titles.

Access to international scientometric databases has become restricted for Russian organisations, and national tools for research assessment have since been developed. In this context, the question of the transparency and reproducibility of journal-selection rules matters a great deal. The aim of this article is to audit the methodology used to construct the international component of the Unified State Register of Scholarly Publication Channels (USRSPC), together with the results of its application to a corpus of international publication channels. We assess the methodology’s internal consistency, the reproducibility of the resulting decisions, and the interpretability of the final journal ranking. The empirical analysis utilises the normalisation of files submitted by the Russian Centre for Scientific Information (RCSI) for review by the divisions of the Russian Academy of Sciences, the merging of duplicate records across subject categories, and the comparison of the 2026 USRSPC recommendations with the international component of the 2023 Whitelist. The findings show that the methodology combines, within a single final ranking, indicators of a journal’s scientific visibility, technical governance, institutional affiliation, geographic profile, and legal and political grounds. Treating these distinct dimensions as one category makes the ranking difficult to interpret and the resulting ranking decisions hard to reproduce. On this basis, we propose revisions to the scoring rules and, more broadly, to the design of the national list of scholarly publication channels.