{
    "matter": "BRT-PAT-2-PROV",
    "document_type": "Patent terminology definitions",
    "definitions": [
        {
            "term": "Accuracy",
            "definition": "“Accuracy” refers to a performance metric indicating the degree to which outputs, classifications, or decisions produced by a system, apparatus, or method correspond to reference, expected, or ground-truth outcomes. Accuracy may be measured as an overall proportion of correct results, as a weighted metric, as a class-specific accuracy, or using related formulations, or the like. In some embodiments, accuracy may be used in training, validation, monitoring, or operational evaluation of a system, apparatus, or method such as for example a screening and/or decision system."
        },
        {
            "term": "Analysis",
            "definition": "“Analysis” refers to a computational process performed on one or more inputs to extract features, detect patterns, identify relationships, generate insights, produce intermediate representations, or inform a subsequent operation or output. Analysis may involve operations such as parsing, comparing, classifying, summarizing, transforming, embedding, clustering, or the like. The analysis may be applied to structured or unstructured data and may be performed by deterministic algorithms, statistical methods, machine learning models, or the like. In the context of artificial intelligence systems, analysis may include processing input data using model parameters to produce an output, such as evaluating a manuscript to produce editorial feedback, or the like. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "artificial intelligence",
            "definition": "Intelligence\" refers to a computational system, module, of the like capable of performing tasks typically requiring human intelligence. These tasks include learning from examples, pattern recognition, decision-making, natural language understanding, and more. AI systems and/or modules can employ a variety of models and techniques, including artificial neural networks (ANNs), machine learning, and deep learning."
        },
        {
            "term": "Artificial Intelligence (AI) system",
            "definition": "“Artificial Intelligence (AI) system” refers to a computational system, module, component, device, or arrangement of hardware, software, firmware, or combinations thereof configured to perform one and/or more tasks that may otherwise be performed by a human, including learning, inference, pattern recognition, prediction, classification, clustering, generation, optimization, planning, control, decision-making, or the like. An AI system may include one and/or more machine-learning models, neural networks, statistical models, rule-based engines, probabilistic models, symbolic reasoning components, or hybrid architectures. Example model types may include convolutional neural networks (CNNs), encoder–decoder architectures, transformer-based models, recurrent neural networks, generative models (including GANs, VAEs, diffusion models, autoregressive models), support-vector machines, decision trees, ensemble methods, Bayesian models, or the like. Synonyms include intelligent system, learning system, machine-learning system, AI module, cognitive system, or the like."
        },
        {
            "term": "Artificial intelligence model",
            "definition": "“Artificial Intelligence (AI) Model” refers to a parameterized computational representation configured, when trained, to map input data to output by learning from data and/or rules. An AI model may perform prediction, classification, segmentation, detection, generation, decision-making, or the like. Examples include, without limitation, neural networks (e.g., convolutional, recurrent, transformer, encoder–decoder including U-Net), generative models (e.g., GAN, VAE, diffusion), probabilistic or margin-based models, and tree/ensemble methods. An AI model may be trained using supervised, semi-supervised, self-supervised, unsupervised, or reinforcement learning and may be deployed alone or as a component of an AI system."
        },
        {
            "term": "Assessment",
            "definition": "\"Assessment\" refers to a process or result of evaluating, analyzing, or appraising a subject matter, which may include evaluating quality, relevance, accuracy, completeness, or compliance with predefined or dynamic criteria. Synonyms for “assessment” may include evaluation, appraisal, analysis, review, scoring, or the like. The assessment may be performed manually by a person, automatically by a computing system, or through a combination of human and automated operations. The subject of the assessment may include data, text, manuscripts, models, system behavior, performance metrics, procedural outcomes, or the like. An assessment may be based on objective criteria, subjective judgment, machine-learned models, rule-based algorithms, or any combination thereof. In the context of machine learning or artificial intelligence systems, the assessment may also involve determining how well a generated output meets a desired goal, aligns with training objectives, or satisfies domain-specific requirements. The assessment may yield a score, ranking, narrative explanation, recommendation, revision suggestion, or the like. The results of the assessment may be stored, communicated, visualized, or used as input to other modules, systems, or processes. The term assessment may be used flexibly in workflows involving document review, automated screening, peer review, content moderation, or the like. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Associate editor",
            "definition": "\"Associate editor\" refers to a person affiliated with a publication entity—such as a journal, academic publisher, or editorial board—who supports or collaborates with one or more editors in managing the evaluation, review, and preparation of content submitted for publication. Synonyms for “associate editor” may include assistant editor, section editor, editorial reviewer, supporting editor, or the like. An associate editor may be assigned to handle submissions within a particular subject area, domain, or section, and may coordinate peer reviews, evaluate reviewer feedback, and make preliminary assessments regarding the quality, relevance, or compliance of the manuscript. The associate editor may provide revision recommendations, suggest editorial comments, or generate preliminary screening output that is reviewed by a supervising editor. In systems involving automated editorial processing, an associate editor may validate, refine, or override machine-generated outputs prior to forwarding decisions or feedback to an editor or author. The associate editor may serve as part of a multi-tiered editorial decision-making structure to ensure academic rigor, ethical compliance, or alignment with publication standards. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Attention",
            "definition": "\"Attention\" refers to a computational mechanism within a neural network—particularly within transformer architectures—that dynamically assigns weights to different elements of an input sequence (e.g., words, tokens, sentences, or whole written works), allowing the model to selectively focus on parts of the input that are most relevant to a given task or context. This mechanism enables the model to capture relationships between distant or contextually significant elements in the data and plays a central role in generating coherent and context-sensitive outputs. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Attention layers",
            "definition": "“Attention layers” refers to one or more computational layers within a neural network that implement an attention mechanism to assign contextual weights to elements of an input sequence. In one embodiment, attention layers are neural network components designed to compute contextual relevance between elements of input data. An attention layer may operate on token embeddings, feature vectors, hidden states, or the like, and compute weighted combinations of values based on learned relationships between query vectors, key vectors, and value vectors, or the like. Attention layers may be configured to perform self-attention, cross-attention, multi-head attention, or the like, and are commonly used in transformer-based architectures to capture dependencies between elements regardless of their position in a sequence. Attention layers may be stacked, combined with feed-forward layers, or integrated into encoder-decoder architectures to support tasks such as text generation, classification, summarization, or the like."
        },
        {
            "term": "Attention mechanism",
            "definition": "\"Attention mechanism\" refers to a computational process, module, apparatus, logic, or the like within an artificial intelligence system or model that selectively emphasizes certain parts of an input sequence based on their relevance to a particular context, task, or prompt. The computational process may include the generation and comparison of vector representations—such as queries, keys, and values—for individual elements in the sequence, followed by the computation of alignment or compatibility scores, such as dot products or scaled dot products, to determine the relative importance of each element. These importance scores may then be normalized (e.g., using softmax) and applied to weight the corresponding values, enabling the model to focus its processing on semantically or contextually significant content. An attention mechanism may be implemented in transformer-based models, encoder-decoder architectures, and other machine learning frameworks. The attention mechanism may facilitate improved handling of long-range dependencies, ambiguity resolution, and contextual coherence in tasks such as language generation, summarization, classification, or the like. Synonyms for “attention mechanism” may include “contextual weighting engine,” “neural attention process,” “relevance alignment layer,” “dynamic focus module,” or the like. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Author",
            "definition": "\"Author\" refers to a person who originates, drafts, or contributes substantively to the creation of a manuscript or other written work intended for publication, distribution, or peer review. Synonyms for “author” may include manuscript creator, contributor, submitting writer, originating writer, content originator, or the like. The author may prepare content in the form of text, figures, tables, citations, or other substantive elements that collectively form the manuscript. The author may be an individual or one of a group of co-authors and may submit the manuscript through an electronic submission system for editorial review or evaluation.\r\n\r\nIn the context of manuscript screening systems, the author may interact with automated systems, editors, or associate editors to receive editorial feedback, editorial comments, screening decisions, or revision requests. The author may respond to editorial comments, provide revised versions, or otherwise participate in iterative refinement of the manuscript. The role of the author may be recognized during training, screening, review, or feedback workflows associated with artificial intelligence models used for editorial evaluation. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Authorship anomalies",
            "definition": "“Authorship anomalies” refers to unusual, inconsistent, or suspicious author-related characteristics observed across manuscripts, submissions, records, or the like. Authorship anomalies may include frequent author name changes, sudden topic shifts inconsistent with prior expertise, mismatched affiliations, mismatched identifiers, repeated identical contributor statements, or similar irregular authorship patterns."
        },
        {
            "term": "Citation relevance",
            "definition": "\"Citation relevance\" refers to the degree to which a citation within a manuscript or other written work meaningfully supports, contextualizes, or relates to the content in which it appears. Synonyms for “citation relevance” may include reference alignment, bibliographic relevance, source appropriateness, contextual citation validity, or the like. Citation relevance may be evaluated based on factors such as the topical similarity between the citation and the citing content, the temporal proximity of the cited work to current knowledge, the credibility or authority of the cited source, or the appropriateness of the citation's placement within the surrounding narrative. Citation relevance may be determined by a human reviewer, a rule-based system, or an artificial intelligence model configured to assess semantic similarity, co-citation frequency, discourse structure, or the like. In the context of automated or AI-assisted quality assessment, citation relevance may be used to ensure that references are not merely present but contribute substantively to the scholarly or technical justification of claims. Assessment of citation relevance may yield descriptive feedback, confidence scores, classification outcomes, or other forms of output that inform editorial decision-making or manuscript review workflows. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Classification accuracy",
            "definition": "“Classification accuracy” refers to a measure of how correctly a computational system assigns items to their intended categories or outcome classes. Classification accuracy may be calculated in various ways, including, without limitation, the proportion of correctly classified samples, weighted or averaged accuracy across multiple classes, accuracy computed relative to a reference dataset containing known or expected labels, or the like. In some embodiments, classification accuracy is used as part of a training objective, performance evaluation process, quality-control mechanism for manuscript screening systems, or the like."
        },
        {
            "term": "Classification boundaries",
            "definition": "“Classification boundaries” refers to one or more decision thresholds, rules, learned parameters, or dividing functions used by a computational model to distinguish among different categories, labels, or outcome classes. Classification boundaries may be explicit (e.g., rule-based or threshold-based) or implicit (e.g., learned by a machine-learning model), and may operate in one or more dimensions of a feature space, embedding space, or probability distribution. In certain embodiments, classification boundaries are used by an artificial-intelligence system to determine how a manuscript or portion of a manuscript should be categorized relative to one or more screening or editorial outcomes, such as 'reject' or 'not reject'."
        },
        {
            "term": "Cohen’s kappa",
            "definition": "“Cohen’s kappa” refers to a statistical agreement metric that measures the degree of agreement between two sets of categorical decisions and/or labels while accounting for agreement that may occur by chance. Cohen’s kappa may be calculated using different variants, weighting schemes, or averaging techniques, or the like, and may be applied to assess agreement between human reviewers, automated systems, hybrid processes, or combinations thereof."
        },
        {
            "term": "Completeness-audit",
            "definition": "“Completeness-audit” refers to an automated or semi-automated process configured to evaluate whether a manuscript contains information that would ordinarily be expected based on the study design, editorial guidelines, or applicable reporting standards. A completeness-audit may examine the presence of disclosure elements, limitation discussions, sample-size justifications, statistical reporting details, or other expected components of scholarly reporting. In some embodiments, a completeness-audit identifies sections that are missing, incomplete, or inconsistent with other reported information."
        },
        {
            "term": "Compliance assessment",
            "definition": "“Compliance assessment” refers to an evaluation, determination, or analysis of whether a manuscript, document, process, data set, or other subject matter conforms to one or more predefined rules, standards, requirements, policies, or the like. A compliance assessment may be performed manually, programmatically, or automatically by a human reviewer, computing system, or combination thereof. In the context of manuscript screening, a compliance assessment may evaluate adherence to formatting guidelines, submission requirements, citation policies, ethical standards, scope of publication, or the like. The assessment may result in a binary outcome, score, ranking, or narrative explanation, or the like. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Computer program product",
            "definition": "\"Computer program product\" refers to a tangible or non-transitory medium or collection of media that stores one or more sequences of instructions, code modules, data structures, models, or configurations that, when executed or interpreted by one or more computing devices, may enable the devices to perform specified operations, processes, or methods. The computer program product may include any suitable form of memory or storage medium, such as magnetic storage, optical storage, solid-state storage, flash memory, or the like, and may be configured to work in distributed or cloud-based systems. A computer program product may facilitate the implementation of software applications, artificial intelligence systems, machine learning models, data processing pipelines, user interfaces, or the like. Synonyms for “computer program product” include “software product,” “code-bearing medium,” “executable program product,” “stored instruction set,” or the like. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "contextual weights",
            "definition": "“Contextual weights” refers to numerical values computed by an attention mechanism or similar model component that represent the relative importance or relevance of elements within an input sequence, based on their relationship to one another in a given context. Contextual weights are typically derived from comparisons between query vectors and key vectors, or the like, and are used to scale corresponding value vectors or features during model inference. These weights allow a model to emphasize or de-emphasize particular elements of the input when generating intermediate representations or outputs, such as in language modeling, classification, summarization, or the like. Contextual weights may vary dynamically across tasks, sequences, or inference steps, and may be computed using dot product attention, additive attention, or the like. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Corpus",
            "definition": "\"Corpus\" refers to a structured or unstructured body, collection, grouping, or aggregation of data, content, content items, informational material, or the like used for computational processing, analysis, retrieval, generation, training, validation, inference, comparison, evaluation, or the like. A corpus may include linguistic content, symbolic content, domain-specific content, documents, records, code samples, transcripts, labels, annotations, metadata, embeddings, structured datasets, tokenized sequences, vectorized representations, or the like and may be organized as raw text, indexed collections, linked structures, graph-based representations, datasets, or coordinated reference sets. A corpus may be curated by a user, assembled from public or proprietary sources, generated synthetically, dynamically constructed, or incrementally updated and may be deployed in training workflows, retrieval workflows, inferential workflows, generative workflows, or real-time processing environments. Synonyms include training corpus, retrieval corpus, language corpus, dataset, data collection, reference collection, information corpus, dataset collection, machine learning dataset, knowledge base, or the like."
        },
        {
            "term": "cosine similarity",
            "definition": "\"Cosine similarity\" refers to a mathematical measure used to quantify the similarity between two non-zero vectors by calculating the cosine of the angle between them in a multi-dimensional space. Synonyms for “cosine similarity” may include angular similarity, vector angle similarity, semantic vector comparison, directional similarity metric, or the like. In the context of natural language processing or artificial intelligence, cosine similarity may be applied to compare vectorized representations—such as sentence embeddings—of textual data, such as machine-generated editorial comments and human-generated editorial comments.\r\n\r\nCosine similarity may yield a value in the range from -1 to 1, where a value closer to 1 indicates greater similarity in direction (i.e., semantic or contextual alignment), a value near 0 indicates orthogonality or irrelevance, and a value near -1 indicates opposition. Cosine similarity may be used during model training to evaluate the alignment between generated and reference outputs or during inference to assess consistency, paraphrasing, or redundancy. The vectors being compared may be derived from neural network embeddings, sentence encoders, transformer-based language models, or the like. Cosine similarity may support training objectives, retrieval systems, clustering, or editorial quality evaluation. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Dataset",
            "definition": "\"Dataset\" refers to a structured and/or unstructured collection, grouping, aggregation, or set of data instances, records, elements, objects, entries, or representations that may be organized according to one or more formats, schemas, models, relational structures, logical associations, or the like. A dataset may be used for training, testing, validation, retrieval, indexing, inference, analysis, generation, transformation, control, monitoring, or other computational and/or operational purposes and may include input data, output data, labeled and/or unlabeled data, paired data instances, linked data, streamed data, dynamically generated data, or the like. Synonyms include data collection, data set, data repository, data grouping, data corpus, information collection, or the like."
        },
        {
            "term": "decoder module",
            "definition": "“Decoder module” refers to a component or set of components within a computing system, machine learning model, artificial intelligence system, software, or the like configured to generate output based on encoded or processed input representations. A decoder module may operate on intermediate data structures such as embeddings, contextual vectors, hidden states, or the like, and may produce output in the form of text, tokens, classifications, or other response formats. In transformer-based and/or encoder-decoder architectures, the decoder module may include attention layers, feed-forward networks, normalization operations, or the like, and may be configured to perform autoregressive generation, token-by-token prediction, or structured output construction. A decoder module may generate narrative textual output, including multiple paragraphs in prose, in response to an input such as a prompt, embedding, or contextual representation. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Disclosure elements",
            "definition": "“Disclosure elements” refers to statements, sections, declarations, or informational components in a manuscript that communicate procedural transparency, ethical compliance, limitations, funding context, or authorship-related information. Disclosure elements may include statements regarding conflicts of interest, ethical approvals, funding sources, contributor roles, data availability, prior publication, or similar disclosures."
        },
        {
            "term": "Discrepancies",
            "definition": "“Discrepancies” refers to inconsistencies, conflicts, deviations, or contradictions identified within or among portions of a manuscript or related materials. Discrepancies may arise between textual statements, numerical values, figures, tables, metadata, or referenced sources, or the like. By way of example, a discrepancy may include conflicting descriptions of methodology, mismatched sample sizes across sections, numerical results that do not match reported statistics, or inconsistencies between reported conclusions and supporting data, or the like."
        },
        {
            "term": "Edit",
            "definition": "\"Edit\" refers to an action or series of actions by which a person, such as an editor or associate editor, alters, corrects, refines, supplements, or otherwise modifies content, including manuscript screening output. Synonyms for “edit” may include revise, amend, annotate, adjust, rewrite, or the like. The edit may involve changes to wording, structure, tone, factual accuracy, formatting, emphasis, or clarity of the output, and may serve purposes such as improving readability, ensuring compliance with publication standards, clarifying points of critique, or aligning machine-generated content with human editorial judgment.\r\n\r\nThe manuscript screening output subject to edit may include textual output generated by an artificial intelligence model—such as machine-generated editorial comments, quality assessments, compliance determinations, or screening decisions—and the edit may reflect editorial oversight, human intervention, or post-processing adjustments within a publication workflow. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Edit feature",
            "definition": "“Edit feature” refers to any software functionality, user interface control, system capability, or workflow mechanism that allows an authorized user to modify, adjust, annotate, override, supplement, or otherwise change content or parameters associated with operation of screening method, system, or apparatus. An edit feature may enable modification of a manuscript screening output, a prompt, a screening prompt, or instruction used to generate an output, editorial guidance text, classification selections, system recommendations, review criteria, or similar elements. In some embodiments, an edit feature supports iterative refinement, permitting a user to make revisions, trigger re-analysis, or request alternative outputs based on the edited content."
        },
        {
            "term": "Editor",
            "definition": "\"Editor\" refers to a person affiliated with a publication entity—such as a journal, publisher, academic institution, or media organization—who exercises professional judgment in evaluating, managing, or modifying written content submitted for publication. Synonyms for “editor” may include publication reviewer, content overseer, manuscript coordinator, editorial authority, or the like. An editor may be responsible for assessing the quality, structure, clarity, relevance, novelty, and conformity of submitted content with applicable guidelines, ethical standards, or formatting requirements. The editor may oversee peer review processes, make acceptance or rejection decisions, provide revision recommendations, or curate final publication content. In some embodiments, the editor may collaborate with associate editors, reviewers, or automated systems to perform screening, review, or quality control functions. The editor may act as a human-in-the-loop in workflows involving machine-generated editorial output, confirming, modifying, or supplementing such output before author notification or publication. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Editorial comment",
            "definition": "“Editorial comment” refers to a written response, observation, critique, suggestion, or assessment relating to a manuscript or other content, typically intended to provide feedback, guidance, or evaluation. An editorial comment may address aspects of the content’s structure, format, tone, approach, methodology, clarity, style, or the like. Editorial comments may be used in peer review, editorial screening, manuscript assessment, or the like, and may be authored at least in part by a person such as an editor or reviewer or generated by a system such as a machine learning model or rule-based engine. An editorial comment may be authored in whole or in part by a human editor, reviewer, or other participant in the review process, or may be generated in whole or in part by a computing system, including a rule-based engine or machine-learning model. Unless expressly stated otherwise, the term “editorial comment” encompasses both human-generated and machine-generated comments, as well as comments collaboratively produced by human and machine systems."
        },
        {
            "term": "Editorial review",
            "definition": "“Editorial review” refers to a process, workflow, or decision-making activity in which one or more authorized reviewers evaluate a manuscript, its associated metadata, or outputs generated by a screening method, apparatus, or system, for purposes including suitability assessment, compliance checking, quality control, or publication decision support. Editorial review may include examining manuscript content, reviewing automated screening outputs, modifying recommendations, entering comments, requesting revisions, forwarding for peer review, or making preliminary accept–reject determinations. Editorial review may be performed by editors, associate editors, editorial assistants, reviewers, or other designated personnel using software tools, automated systems, methods, apparatuses, or a combination thereof. As used herein, ‘editorial review’ refers to and includes any form of manuscript evaluation, assessment, screening, triage, vetting, quality control, or decision support, whether performed by a human, an artificial intelligence system, or a combination thereof, and whether occurring before, during, or after peer review."
        },
        {
            "term": "electronic submission system",
            "definition": "\"Electronic submission system\" refers to a computing platform, software application, web-based portal, or automated interface configured to receive, transmit, or manage the submission of digital content from users to a centralized processing or review system. Synonyms for “electronic submission system” may include manuscript intake system, digital upload portal, online submission interface, content submission platform, or the like. The electronic submission system may enable authors to submit manuscripts, supporting documents, metadata, or other related content for purposes such as review, publication, compliance evaluation, or editorial screening.\r\n\r\nIn the context of artificial intelligence–assisted manuscript evaluation, the electronic submission system may interface with screening modules, data storage systems, editorial workflows, or notification services. The electronic submission system may include components to validate document formats, authenticate users, associate submissions with user accounts, log submission timestamps, or route manuscripts to appropriate evaluators or automated tools. Submissions handled by the electronic submission system may be used as inputs to artificial intelligence models trained on editorial comment corpora or used for downstream quality, compliance, or similarity assessments. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Embedding space",
            "definition": "“Embedding space” refers to a numerical, vector-based, or otherwise representational space in which inputs — including words, sentences, documents, features, portions, sections, entities, or the like — are mapped to representations that capture contextual, relational, semantic characteristics and/or the like. An embedding space may be produced by a machine-learning model, mathematical transformation, statistical process, or the like, and may exist in one or more dimensions. Relationships such as distances, angles, clustering patterns, neighborhoods, or the like within the embedding space may be used to support similarity analysis, retrieval, classification, recommendation, model training, or related computational tasks."
        },
        {
            "term": "Evidentiary support",
            "definition": "“Evidentiary support” refers to the data, analytical results, citations, or reasoning that substantiate a claim made in a manuscript. Evidentiary support may include statistical results, experimental outcomes, prior-published research, empirical observations, analytical derivations, or the like. Evidentiary support may appear in figures, tables, narrative text, appendices, or supporting documentation."
        },
        {
            "term": "F-1 score",
            "definition": "“F-1 score” refers to a performance metric that represents a combined or harmonic relationship between precision and recall, or between comparable component measures, to summarize classification effectiveness in a single value. The F-1 score may be calculated using mathematical formulas, weighted variations, averaged forms, or related approaches, or the like, and may be applied at the level of individual classes, groups of samples, or entire datasets."
        },
        {
            "term": "Input",
            "definition": "\"Input\" refers to any data, signal, instruction, or other information received or ingested by a system, device, software module, or artificial intelligence model for the purpose of processing, analysis, transformation, or storage. Synonyms for “input” may include received data, incoming data, system input, user-provided data, or the like. Input may originate from a user, device, sensor, database, file, network resource, or another software or hardware component. The input may be in the form of natural language text, numerical values, images, audio signals, structured records, tokenized sequences, embeddings, encoded formats, or the like. Input may be submitted manually, generated automatically, or retrieved programmatically, and may be subjected to preprocessing, formatting, normalization, or tokenization prior to further handling. In the context of artificial intelligence systems or manuscript review workflows, input may include a manuscript, submission metadata, editor comments, editor edits, editorial prompts, reviewer instructions, system parameters, or other contextual information relevant to generating outputs."
        },
        {
            "term": "Limitation discussion",
            "definition": "“Limitation discussion” refers to narrative text in which the authors describe weaknesses, constraints, assumptions, or uncertainties associated with their study. A limitation discussion may address matters such as sample size limits, potential biases, measurement constraints, confounding factors, generalizability restrictions, or similar study limitations."
        },
        {
            "term": "Linguistic claims",
            "definition": "“Linguistic claims” refers to statements, assertions, or conclusions expressed in a manuscript, for example the narrative text, that communicate what the authors believe the results show. Linguistic claims may occur in an Abstract, Discussion, Title, Conclusion, Cover Letter, or similar section. Such claims may reference causation, strength of evidence, certainty, generalizability, novelty, perceived importance of the findings, or the like."
        },
        {
            "term": "Linguistic strength",
            "definition": "“Linguistic strength” refers to the degree of confidence, certainty, emphasis, or assertiveness expressed in a linguistic claim. Linguistic strength may be reflected through wording that conveys definitiveness, probability, likelihood, or limitation. Examples include statements such as “proves,” “demonstrates,” “is strongly associated,” “may suggest,” or similar expressions indicating relative confidence or caution."
        },
        {
            "term": "Machine-generated editorial comment",
            "definition": "“Machine-generated editorial comment” refers to an editorial comment produced by a computing system or software system, which may include one or more trained artificial intelligence models. The editorial comment may be generated in response to analysis of a manuscript. The editorial comment may be generated without direct human authorship and may include suggestions, critiques, or other narrative text intended to address or evaluate aspects of the manuscript’s content, structure, format, tone, approach, methodology, clarity, style, or the like. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Manuscript",
            "definition": "\"Manuscript\" refers to a document or piece of writing that is typically in its original, unprinted, unpublished form, such as a draft of a book, an article, a journal article, scientific research findings, a thesis, a report, a script, or the like. The manuscript may be created by an individual, referred to as an author, or a group of individuals, referred to as authors, and may be in digital or physical format. The content of the manuscript may cover a wide range of topics, including but not limited to, scientific research, literature, history, technology, arts, or any other field of knowledge. The manuscript may be subject to editing, reviewing, and/or revising before it is finalized for publication or presentation. The manuscript may also serve as an input or source material for various processes, such as translation, transcription, summarization, critique, or machine learning model training. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Manuscript screening outcomes",
            "definition": "“Manuscript screening outcomes” refers to the possible results, statuses, or dispositions that may be assigned to a manuscript during an editorial screening process. Examples of manuscript screening outcomes include, without limitation, rejection, no reject, acceptance for further review, request for revision, deferral, referral, or other editorial dispositions indicating whether or how the manuscript will proceed in a review workflow. Manuscript screening outcomes may be generated by human reviewers, automated systems, or a combination thereof."
        },
        {
            "term": "Manuscript screening output",
            "definition": "“Manuscript screening output” refers to one or more machine-generated outputs resulting from automated analysis of a manuscript, the output may include a screening decision (e.g., REJECT, NOT REJECT, REVISE, or the like), a manuscript summary, an objective editorial criteria summary, a subjective editorial criteria, a narrative explanation, a recommendation, editorial guidance, or the like. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Manuscript summary",
            "definition": "“Manuscript summary” refers to a condensed textual representation of the content of a manuscript, intended to capture its main ideas, contributions, findings, conclusions, or the like. A manuscript summary may be generated by a person or by a computing system, and may take the form of a paragraph, abstract, bullet list, or other textual format. The summary may be used to facilitate editorial review, indexing, searchability, screening, or comprehension of the manuscript's subject matter. A manuscript summary may be derived from analysis of the manuscript’s title, abstract, body text, figures, tables, citations, or the like. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Metadata inconsistencies",
            "definition": "“Metadata inconsistencies” refers to mismatches or irregularities in structured information associated with a manuscript submission. Metadata inconsistencies may include conflicting submission dates, mismatched author affiliations, inconsistent correspondence addresses, unusual revision timelines, conflicting declarations, or similar irregular metadata characteristics."
        },
        {
            "term": "Methodological-reporting mismatch",
            "definition": "“Methodological-reporting mismatch” refers to a discrepancy between the methods described in a manuscript and the results that are actually reported. A methodological-reporting mismatch occurs when a study design implies analysis or data that does not appear, or when reported outcomes are inconsistent with the described procedures. Examples include manuscripts describing longitudinal follow-up but presenting only cross-sectional data, promising subgroup analyses that never appear, or reporting conclusions inconsistent with the stated methodology."
        },
        {
            "term": "Misclassification error",
            "definition": "“Misclassification error” refers to any discrepancy, deviation, or incorrect assignment in which a system, model, classifier, decision engine, or the like assigns an input to a label, category, or outcome that differs from an intended, expected, or reference label. Misclassification error may be expressed as a count, percentage, rate, loss contribution, penalty value, performance metric, or the like, and may apply to individual samples, subsets of samples, or entire datasets. In some embodiments, misclassification error is used to guide or evaluate training, optimization, quality control, or operation of artificial-intelligence systems, rule-based systems, decision pipelines, or the like."
        },
        {
            "term": "Model",
            "definition": "\"Model\" refers to an abstract, representational structure that may be implemented in a computational environment to simulate, analyze, predict, generate, or recognize features of objects, systems, data, or processes. In artificial intelligence and machine learning contexts, the model may include a trained parameter space and inference logic configured to perform classification, regression, generation, transformation, or the like, based on input data. Synonyms may include computer model, predictive model, simulation model, artificial intelligence model, or the like. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Model-generated editorial comment",
            "definition": "“Model-generated editorial comment” refers to an editorial comment produced by a trained machine learning model, such as a transformer-based neural network. The editorial comment may be based on a manuscript or input derived therefrom. The editorial comment may be generated without direct human authorship and may include suggestions, critiques, or other narrative text intended to address or evaluate aspects of the manuscript’s content, structure, format, tone, approach, methodology, clarity, style, or the like. The comment is typically the model’s output in response to a prompt or internal decision process and may be evaluated, compared, or refined during model training or tuning. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Module",
            "definition": "\"Module\" refers to a functionally distinct component or logical unit within a system, apparatus, method, or software solution that is configured to perform one or more specific operations. Synonyms for “module” may include component, subsystem, unit, segment, engine, block, or the like. A module may be implemented in hardware, software, firmware, or any combination thereof. A module may be embodied as a software routine, class, object, process, or service, or as a hardware circuit, integrated chip, programmable logic component, or the like. A module may operate independently or in cooperation with one or more other modules and may be configured to receive, process, generate, transmit, or store data, or the like. A module may be distributed across computing environments or reside within a single computational entity."
        },
        {
            "term": "narrative explanation",
            "definition": "“Narrative explanation” refers to a sequence of text expressed in natural language that conveys reasoning, justification, interpretation, or insight in a descriptive or explanatory form. A narrative explanation may consist of one or more sentences or paragraphs written in prose and may include summaries, evaluations, recommendations, or the like. It may be directed toward clarifying a decision, describing an observation, explaining a process, or supporting an assessment, or the like. A narrative explanation may be generated by a human author or by a computing system, such as a machine learning model. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Numerical embeddings",
            "definition": "\"Numerical embeddings\" refer to the representation of data, such as words, phrases, ideas, sentences, or even larger units of text, as vectors of real numbers in a high-dimensional space. These embeddings may be generated by or for machine learning models, such as neural networks, and may be used to capture semantic or syntactic relationships between the data elements. The numerical embeddings may be generated without direct human intervention and may include information about the data's context, relationships, frequency, or other characteristics. The embeddings are typically the model's output in response to a given input and may be evaluated, compared, or refined during model training or tuning. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Omissions",
            "definition": "“Omissions” refers to missing, incomplete, or omitted information that would ordinarily be expected to appear in view of the subject matter being discussed, applicable reporting standards, or typical editorial expectations. Omissions may relate to missing methodology details, incomplete description of results, missing disclosures, absent data tables, unreported statistical assumptions, or the like. In certain embodiments, omissions include situations where required or expected content is referenced elsewhere but not actually presented, or the like."
        },
        {
            "term": "Output",
            "definition": "“Output” refers to any signal, data, result, action, or information produced, emitted, displayed, transmitted, or stored by a system, device, software program, or computational process. Output may be directed to a person, machine, system component, storage medium, communication interface, or the like, and may be presented in various forms, including textual output, graphical output, auditory signals, binary data, structured records, or the like. Output may result from one or more internal operations or computations performed by a machine learning model, algorithm, processor, or the like, and may be used for presentation, logging, control, further analysis, or feedback. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Paper-mill generated manuscripts",
            "definition": "“Paper-mill generated manuscripts” refers to manuscripts produced, sold, or distributed by entities that generate publications or authorship opportunities for profit rather than legitimate scholarship. Paper-mill manuscripts may display repeated templates, reused figures, generic methods sections, fabricated data, recycled phrasing, or other systematic artifacts of mass-produced submissions."
        },
        {
            "term": "Portion",
            "definition": "“Portion” refers to a subset, segment, component, or division of a larger whole, which may be defined by position, content, function, structure, or the like. A portion may be continuous or discontinuous, fixed or variable in size, and may include one or more elements of a larger data structure, physical object, or conceptual entity. For example, in the context of text or documents, a portion may include a word, phrase, sentence, paragraph, section, or the like. A portion may be selected, referenced, analyzed, or processed independently or in combination with other portions. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Precision",
            "definition": "“Precision” refers to a performance metric indicating the proportion, fraction, or rate of predicted positive outcomes that correspond to correct or reference-validated positive outcomes. Precision may be calculated in different forms, may be weighted or averaged across multiple categories, and may be used as part of broader evaluation, optimization, or decision-making processes."
        },
        {
            "term": "Processor",
            "definition": "\"Processor\" refers to any electronic circuitry, component, chip, die, package, or module that may be configured to receive, interpret, decode, and perform machine-executable instructions. The processor may perform arithmetic, logical, control, or data processing operations, and may support general-purpose or application-specific computation. The processor may include or be implemented as a central processing unit (CPU), graphics processing unit (GPU), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), digital signal processor (DSP), system-on-chip (SoC), virtual processor, processor core, or the like. The processor may be implemented in hardware, firmware, software, or any combination thereof, and may operate in standalone, distributed, or cloud-based environments. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Prompt",
            "definition": "\"Prompt\" refers to an input signal, message, data structure, or the like provided to a computational system—such as an artificial intelligence system or large language model (LLM)—that initiates, guides, or influences the system’s generation, classification, or analysis of output. Synonyms for \"prompt\" include query, input query, user prompt, task directive, system instruction, or the like.\r\n\r\nA prompt may be expressed in natural language, code, structured syntax, formatted input, or embedded data representations, and may include one or more instructions, questions, or contextual inputs. The prompt may be static, dynamically generated, user-defined, or derived from prior interactions, system state, or metadata. In the context of machine learning and generative artificial intelligence, a prompt may shape or constrain the model’s inference behavior, including output tone, content domain, formatting, or level of specificity. Prompts may also be used during training, evaluation, or fine-tuning of models to simulate realistic tasks, enforce structure, or provide reference patterns, or the like."
        },
        {
            "term": "Publisher guidelines",
            "definition": "\"Publisher guidelines\" refers to a set of formal and/or informal instructions, protocols, policies, rules, standards, formatting requirements, editorial criteria, or the like established or adopted by a publishing entity, editorial board, or dissemination platform that may govern the submission, review, formatting, ethics, authorship, citation, or dissemination of manuscripts or other content. Synonyms for \"publisher guidelines\" include \"submission criteria\", \"editorial standards\", \"publication requirements\", \"formatting rules\", and \"editorial policies\". The publisher guidelines may vary between different publishers or publications and may evolve over time. The publisher guidelines may include explicit or implied expectations regarding style, tone, length, citation format, originality, conflict of interest disclosures, ethics disclosures, data sharing, image quality, or the like. The publisher guidelines may be expressed in natural language, structured metadata, machine-readable formats, or combinations thereof. The publisher guidelines may be accessed manually by a human or programmatically by a system and may serve as reference standards for screening, evaluating, or formatting submissions (e.g. manuscripts) in editorial, academic, legal, or commercial contexts. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Quality assessment",
            "definition": "\"Quality assessment\" refers to a form of assessment that evaluates the degree to which a subject, such as a manuscript, system output, data set, model result, or process outcome, satisfies predefined or context-sensitive standards of quality. Synonyms for “quality assessment” may include quality evaluation, quality review, performance appraisal, standard compliance check, or the like. A quality assessment may be based on one or more criteria such as clarity, completeness, accuracy, consistency, originality, readability, technical soundness, recency criteria, or alignment with domain-specific norms or publisher guidelines. The criteria used may be fixed, adaptive, or learned from data. A quality assessment may be conducted by a human evaluator, an automated system, an artificial intelligence module, or a combination thereof. In artificial intelligence systems, a quality assessment may include computing similarity metrics, analyzing coherence, fluency, or informativeness of generated text, or determining whether the output meets specified objectives. The results of a quality assessment may include a rating, score, rank, flag, editorial comment, or the like, and may inform subsequent decision-making or further processing steps. The quality assessment may be domain-specific and applied within workflows involving content moderation, peer review, machine-generated text evaluation, or other review processes. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Recall",
            "definition": "“Recall” refers to a performance metric indicating the proportion, fraction, or rate of actual positive outcomes that are correctly identified by a system, classifier, apparatus, method, or model. Recall may be computed using different formulations, may be applied to one or more categories, and may be incorporated into composite metrics, optimization procedures, evaluation frameworks, or the like."
        },
        {
            "term": "recency criteria",
            "definition": "\"Recency criteria\" refers to one or more evaluation parameters used to assess the timeliness or up-to-date nature of content in relation to a relevant temporal context. Synonyms for “recency criteria” may include currency standards, timeliness measures, temporal relevance checks, freshness indicators, or the like. Recency criteria may be applied to textual content, citations, references, data sources, methodologies, or any other element within a manuscript, data set, or model output. The recency criteria may evaluate whether cited materials or referenced works fall within an acceptable time window, whether newer or more authoritative sources are omitted, or whether the subject matter reflects developments within the field. The criteria may be defined statically by rules or thresholds, dynamically based on publication dates and topical shifts, or learned through artificial intelligence models trained on reviewer behavior or editorial guidance. The application of recency criteria may be used in workflows related to academic publishing, machine-generated content review, or automated peer assessment, and may yield binary outcomes, scores, ranked outputs, narrative comments, or the like. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "record",
            "definition": "\"Record\" refers to a structured or unstructured representation of data associated with a particular instance, example, observation, or the like within a dataset. A record may include any combination of textual, numerical, categorical, or metadata elements related to a subject of interest. In one embodiment, a record may comprise a digital object such as a JSON object, database row, a tuple of data, a serialized file, or the like. A record may store information in key-value pairs, tabular fields, tuples, nested formats or the like, and may represent a training example, inference input, user interaction, or the like. A record may be used to organize inputs, outputs, labels, or annotations for use in machine learning, data analysis, or decision-support systems. Synonyms for 'record' may include 'data instance,' 'training example,' 'data item,' 'entry,' or the like."
        },
        {
            "term": "Reported results",
            "definition": "“Reported results” refers to findings, outcomes, data summaries, or analytical conclusions that are expressly presented within a manuscript. Reported results may include narrative descriptions, tables, figures, model outputs, statistical summaries, conclusions derived from analysis, or the like. Reported results may appear in a Results section, Abstract, Discussion, figures, supplementary files, related materials, or the like."
        },
        {
            "term": "Reporting omission",
            "definition": "“Reporting omission” refers to a type of omission in which a variable, parameter, outcome, or other element described in one portion of a manuscript is not correspondingly reported, analyzed, or discussed in another portion where it would ordinarily be expected. A reporting omission may include, for example, a variable listed in the Methods section that lacks a corresponding summary in the Results section, an outcome referenced in the Abstract but not reported in the body of the manuscript, or a statistical test described but not presented, or the like."
        },
        {
            "term": "Review",
            "definition": "\"Review\" refers to a process or set of actions by which submitted content, such as a manuscript, is examined, evaluated, or assessed for purposes such as quality control, compliance, relevance, completeness, originality, or suitability for publication. Synonyms for “review” may include evaluation, examination, critique, editorial analysis, inspection, vetting, or the like. The review may be performed manually by one or more individuals, such as editors or peer reviewers, or automatically by computing systems, including artificial intelligence models trained to simulate and/or support editorial decision-making.\r\n\r\nThe review may include operations such as identifying strengths and weaknesses, applying scoring metrics, generating editorial comments, checking adherence to publisher guidelines, or determining whether the manuscript should proceed to further stages such as peer review, revision, acceptance or the like. In an artificial intelligence–assisted workflow, the review may further include processing manuscript content using machine learning techniques, generating a manuscript screening output, or comparing outcomes to historical or user-defined editorial standards. As used herein, ‘Review’ may also be referred to as, or encompass, evaluation, assessment, screening, triage, vetting, quality control, editorial examination, or the like."
        },
        {
            "term": "revision recommendations",
            "definition": "\"Revision recommendations\" refers to one or more suggestions, proposals, or directives intended to guide modification, correction, or improvement of a manuscript or other written content. Synonyms for “revision recommendations” may include edit suggestions, improvement directives, content adjustment proposals, manuscript change indicators, or the like. Revision recommendations may address structural organization, logical flow, language clarity, grammar, citation usage, methodology, tone, formatting, compliance with publisher guidelines, or other aspects of the manuscript. The revision recommendations may be generated manually by human reviewers or editors, or automatically by an artificial intelligence model, natural language processing engine, rule-based system, or a combination of these. Such recommendations may be expressed as discrete edit commands, annotated text, inline commentary, summary statements, or structured reports, and may be delivered in natural language, markup format, user interface components, or the like. Revision recommendations may vary in specificity and may be used as part of peer review workflows, automated pre-submission checks, editorial quality assessments, or iterative manuscript refinement processes. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Rogue-author behavior",
            "definition": "“Rogue-author behavior” refers to conduct by an individual or group intended to manipulate, circumvent, or exploit the scholarly publication process. Rogue-author behavior may include fabricated data, false authorship declarations, purchased authorship, ghostwritten submissions, undisclosed conflicts, plagiarism, or similar academic misconduct."
        },
        {
            "term": "Rule-based manuscript screening systems",
            "definition": "“Rule-based manuscript screening systems” refers to systems, engines, workflows, software tools, or the like that evaluate manuscripts using predetermined rules, heuristics, logical conditions, thresholds, or procedural checks rather than (or in combination with) statistical learning approaches. Such rules may evaluate formatting compliance, citation style, word count limits, missing sections, disclosure completeness, language patterns, metadata consistency, or other editorial conditions, or the like. In some embodiments, rule-based manuscript screening systems operate independently, while in other embodiments they operate alongside or in combination with machine-learning systems."
        },
        {
            "term": "Sample-size justification",
            "definition": "“Sample-size justification” refers to an explanation, calculation, rationale, or discussion regarding how the sample size of a study was determined or why it is appropriate for the intended analyses. Sample-size justification may include power calculations, effect-size assumptions, feasibility explanations, economic constraints, references to prior-published benchmarks, or the like."
        },
        {
            "term": "Screening",
            "definition": "\"Screening\" refers to a process by which an input—such as a manuscript, document, application, proposal, or the like—is evaluated against one or more criteria to determine whether the input satisfies, violates, deviates from, or otherwise relates to defined thresholds, standards, or expectations. Synonyms for “screening” may include evaluating, assessing, reviewing, examining, filtering, or the like. Screening may be performed by a human, a machine, or a combination thereof, and may involve qualitative, quantitative, heuristic, statistical, rule-based, or machine learning-based processes, or the like.\r\n\r\nIn the context of manuscript review workflows, screening may include the automated or semi-automated analysis of submitted manuscripts to determine suitability for publication, adherence to publisher guidelines, quality of writing, originality of content, relevance of citations, or other editorial standards. Screening may result in a screening decision (e.g., reject or not reject), generation of one or more editorial comments, routing of the manuscript for further human review, and/or the like. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Screening Decision",
            "definition": "“Screening Decision” refers to the preliminary editorial determination made to assess whether a submitted manuscript meets certain criteria for further consideration. The determination may be made by a journal or editorial team with or without the assistance of software and/or computer systems. The screening decision may be based on subjective criteria (e.g., perceived novelty, clarity of presentation, and editorial fit) and/or objective criteria (e.g., adherence to formatting rules, completeness of statistical reporting, detection of plagiarism, confirmation of required ethical disclosures, and validation of authorship metadata). The criteria may include scope, relevance, content quality (clarity, methodology), novelty, research relevance, research timeliness, academic rigor, compliance with submission guidelines such as ethical requirements and study requirements, and/or other editorial standards. The Screening Decision may and typically does occur before formal peer review and may result in immediate rejection, acceptance for review, requests for modification, or the like. The term encompasses a range of synonymous or closely related practices, including editorial triage, desk decisions, gatekeeping determinations, and initial suitability reviews. In certain embodiments, a Screening Decision is a preliminary decision made as part of a larger manuscript review process for a given manuscript. In another embodiment, is a final decision made a given manuscript, which may or may not be reviewed by a human editor before notice of the Screening Decision is sent to the author. A screening decision may be generated in whole or in part by a human editor, associate editor, or reviewer, or may be generated in whole or in part by a computing system, including a rule-based engine or machine-learning model. The term “screening decision” encompasses both human-generated and machine-generated screening determinations unless expressly stated otherwise."
        },
        {
            "term": "Screening prompt",
            "definition": "\"Screening prompt\" refers to a signal, instruction, input string, or command used to initiate or guide the processing behavior of a computing system, including artificial intelligence systems and large language models. Synonyms for “screening prompt” may include input prompt, model directive, query string, evaluation instruction, or the like. The screening prompt may include text, code, metadata, parameters, or any data construct interpretable by the system to trigger specific output behavior. The screening prompt may be static or dynamically generated, may include system-level or user-level context, and may operate alone or in combination with additional inputs to shape model response behavior.\r\n\r\nIn the context of screening a manuscript using a large language model (LLM), a screening prompt may serve to direct the LLM to analyze a manuscript for editorial or evaluative purposes. The screening prompt may instruct the model to identify strengths or weaknesses, assess formatting or clarity, determine compliance with publisher guidelines, generate editorial comments, generate manuscript screening output, or the like. The screening prompt may encode a particular tone, level of detail, or criteria against which the manuscript is to be evaluated, and may be based on publisher requirements, editorial policies, or reviewer expectations, or the like. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Semantic similarity",
            "definition": "“Semantic similarity” refers to a measure of how closely two items — such as texts, phrases, vectors, annotations, or the like — correspond in meaning, informational content, or intended interpretation, even where their wording or structure differs. Semantic similarity may be computed or assessed using natural-language processing techniques, statistical models, similarity functions, embedding-based comparisons, human evaluation, combinations thereof, or the like. In some embodiments, semantic similarity is used to compare machine-generated content to human-generated content, evaluate consistency across systems, or support training or optimization processes, or the like."
        },
        {
            "term": "Sequence",
            "definition": "\"Sequence\" refers to an ordered set of related elements, such as numbers, characters, words, strings, sentences, paragraphs, pages, tables, images, or events, arranged in a specific pattern or following one or more rules. The sequence may be generated by a mathematical algorithm, a machine learning model, software, a physical process, or other means. The elements in the sequence may be discrete or continuous, finite or infinite, and may represent data, instructions, words, sentences, concepts, ideas, paragraphs, pages, states, transformations, or other types of information. The sequence may be used for computation, communication, control, analysis, prediction, representation, or other purposes. The order, structure, properties, and relationships of the elements in the sequence may be subject to study, manipulation, optimization, or other operations. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "similarity",
            "definition": "“Similarity” refers to a quantitative or qualitative measure indicating a degree of relatedness, correspondence, resemblance, alignment, or the like between two items, such as two texts, vectors, features, model outputs, or the like. Similarity may be assessed using numerical metrics (e.g., cosine similarity, correlation, distance-based measures, or the like) or rule-based or semantic comparison techniques. In certain embodiments, similarity is used to compare machine-generated content to human-generated content, to evaluate consistency of outputs across models, or to guide training or optimization processes."
        },
        {
            "term": "Similarity score",
            "definition": "\"Similarity score\" refers to a numerical value that represents the degree of relatedness, correspondence, or alignment between two data representations, often computed using a defined similarity metric. Synonyms for “similarity score” may include semantic match value, relatedness measure, alignment coefficient, similarity index, or the like. The similarity score may quantify the extent to which two textual, visual, numerical, or other data inputs are comparable in meaning, structure, content, or context.\r\n\r\nIn the context of comparing editorial comments, the similarity score may be derived from applying cosine similarity to vector representations of a machine-generated editorial comment and a human-generated editorial comment for the same manuscript. The score may range between a minimum and maximum boundary defined by the similarity metric, such as 0 to 1 for cosine similarity, and may be used to guide or evaluate training objectives for an artificial intelligence module. A higher similarity score may indicate greater semantic overlap or conceptual equivalence, whereas a lower similarity score may indicate divergence or dissimilarity. The similarity score may be used in supervised learning, fine-tuning, model evaluation, or feedback loops to refine output quality. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Statistical report",
            "definition": "“Statistical report” refers to information presented in a manuscript that communicates the results of statistical analyses or computations. A statistical report may include, for example, p-values, confidence intervals, effect sizes, test statistics, model coefficients, variance estimates, descriptive tables, or the like. A statistical report may appear in text, figures, tables, supplementary materials, associated metadata, or the like."
        },
        {
            "term": "Statistical reporting details",
            "definition": "“Statistical reporting details” refers to numerical, descriptive, or analytical information that clarifies how statistical analyses were conducted or interpreted. Statistical reporting details may include confidence intervals, test statistics, variance measures, assumptions checked, correction procedures, model specifications, or similar components of statistical description."
        },
        {
            "term": "Statistical results",
            "definition": "“Statistical results” refers to outcomes, estimates, measurements, or analytical findings produced through the application of statistical methods to a dataset. Statistical results may include values such as means, proportions, regression coefficients, p-values, confidence intervals, effect sizes, test statistics, or probability estimates. Statistical results may be expressed in narrative text, tables, figures, computational outputs, or the like."
        },
        {
            "term": "Study design",
            "definition": "“Study design” refers to the overall structure, plan, methodology, or framework under which a research study is conducted. Study design may include choices regarding sample selection, control conditions, randomization strategies, timing of measurements, statistical testing approach, or data collection methodology. Examples of study designs include randomized controlled trials, observational studies, longitudinal studies, case–control studies, cross-sectional studies, or similar research frameworks."
        },
        {
            "term": "Textual output",
            "definition": "“Textual output” refers to one or more sequences of characters, symbols, or encoded representations that convey information in written language form. Textual output may include words, phrases, sentences, paragraphs, sections, or the like, and may be formatted in plain text, markup, structured metadata, or other representations suitable for display, storage, or further processing. In certain contexts, textual output may comprise a narrative composed of multiple paragraphs written in prose, such as summaries, explanations, evaluations, recommendations, or the like. Textual output may be generated by a computing system, machine learning model, rule-based engine, or the like, and may be directed to human or machine recipients. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "Token sequences",
            "definition": "\"Token sequences\" refers to ordered collections of tokens produced during the tokenization of input data, such as text, code, or symbolic content, where each token represents a discrete segment of the original input. Synonyms for \"token sequences\" include token streams, lexical sequences, parsed units, encoded input representations, or the like.\r\n\r\nA token sequence may preserve the syntactic, semantic, or contextual structure of the original input and may include one or more tokens that correspond to words, sub words, punctuation marks, or other linguistic or symbolic elements. Token sequences may be represented as arrays, lists, or tensors and may be further transformed into embeddings or vectorized representations for use by artificial intelligence systems, such as large language models, transformer-based networks, neural encoders, or other machine learning pipelines. Token sequences may vary in length and structure depending on the input content, tokenization strategy, language, or model-specific vocabulary, and may include special tokens such as padding tokens, classification tokens, or separators that guide downstream processing. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Tokenization",
            "definition": "\"Tokenization\" refers to a computational process that segments a stream of input data—such as natural language text—into smaller units called tokens, which may be words, sub words, characters, punctuation marks, semantic elements, or symbolic representations suitable for downstream processing. Synonyms for \"tokenization\" include lexical segmentation, text parsing, linguistic decomposition, sub word encoding, or the like.\r\n\r\nTokenization may be performed using rule-based, statistical, or learned approaches, and may involve fixed or variable-length segmentation strategies. Tokenization may operate in conjunction with vocabulary constraints, language-specific heuristics, byte-level encoding techniques, or sub word encoding algorithms such as byte-pair encoding (BPE), WordPiece, or SentencePiece. The tokenization process may include or be preceded by normalization procedures such as case folding, punctuation removal, whitespace trimming, or Unicode canonicalization. Tokenization enables artificial intelligence systems—including large language models, transformer-based models, and other neural network architectures—to represent text in structured or numerical form, thereby facilitating further analysis, embedding, attention modeling, or generation tasks. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "Training objective",
            "definition": "\"Training objective\" refers to a defined goal, criterion, or set of conditions used to guide the learning process of an artificial intelligence model, machine learning system, or neural network during training. Synonyms for “training objective” may include optimization goal, learning criterion, loss function target, model supervision directive, or the like. The training objective may quantify how well the model performs a task and may direct the adjustment of model parameters to minimize or maximize a defined measure of performance. In some embodiments, the training objective may incorporate one or more performance metrics, similarity comparisons, classification accuracy measures, or quality-based evaluations.\r\n\r\nIn the context of editorial content generation, the training objective may include comparing a machine-generated editorial comment for a manuscript to a human-generated editorial comment for the same manuscript using a similarity metric such as cosine similarity, thereby encouraging the model to produce outputs that align semantically with human-provided output. The training objective may also include additional components such as language fluency, coherence, informativeness, or compliance with editorial guidelines. The training objective may be used in supervised, semi-supervised, or reinforcement learning workflows and may be updated or tuned over time to reflect evolving editorial standards, reviewer preferences, or corpus characteristics. \r\nIn certain embodiments, a composite training objective may be employed. A composite training objective refers to a training objective that includes two or more individual objective components that are optimized jointly, such as a combination of classification accuracy, semantic similarity, language quality, or rule-based compliance. The individual components of a composite training objective may be combined through weighting, aggregation, or other optimization strategies, and may be adjusted over time to balance competing editorial goals or performance considerations."
        },
        {
            "term": "Training pairs",
            "definition": "“Training pairs” refers to a collection of two-part data structures used to train a machine learning model. Each pair comprises a first component (such as an input, stimulus, or query) and a second component (such as a corresponding label, target output, response, feedback, or the like), which are associated for the purpose of supervised, semi-supervised learning, or the like. Training pairs may be used to teach or train a model to associate certain types of inputs with corresponding outputs by adjusting internal parameters to minimize error between predicted and actual outputs. Examples include, but are not limited to, an image and its category label, a question and its corresponding answer, or a manuscript and a human-generated editorial comment. Training pairs may be represented in structured or unstructured form, may include metadata, and may be stored in databases, files, or serialized objects. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)"
        },
        {
            "term": "training records",
            "definition": "“Training record” or “Training records” refers to a collection of two-part data structures used to train a machine learning model. Each record comprises a first component (such as an input, stimulus, or query) and a second component (such as a corresponding label, target output, response, feedback, or the like), which are associated for the purpose of supervised, semi-supervised learning, or related machine-learning processes. Training records may be used to teach or train a model to associate certain types of inputs with corresponding outputs by adjusting internal parameters to minimize error between predicted and actual outputs. Examples include, but are not limited to, an image and its category label, a question and its corresponding answer, or a manuscript and a human-generated editorial comment. Training records may be represented in structured or unstructured form, may include metadata, and may be stored in databases, files, or serialized objects."
        },
        {
            "term": "Unsupported claims",
            "definition": "“Unsupported claims” refers to statements, assertions, or conclusions that extend beyond, conflict with, or are insufficiently supported by the data, analysis, or evidence presented in a manuscript. Unsupported claims may include overstating causation, exaggerating effect size, generalizing beyond the scope of the study population, drawing conclusions not tested by the reported analysis, or the like. In some embodiments, unsupported claims include claims lacking citations, claims based solely on speculation, or claims contradicted by data elsewhere in the manuscript, or the like."
        },
        {
            "term": "User-defined corpus",
            "definition": "“User-defined corpus” or “Human-defined corpus” refers to a dataset that includes training data records selected, curated, or otherwise designated or prepared by a user and/or system administrator. Each training record may include a human generated/authored manuscript by a third-party an author, a corresponding human-generated/authored editorial comment produced for that manuscript, and and/or a corresponding human-generated/authored screening decision produced for that same manuscript. In such one an embodiment, the manuscript may serve as the input for two labels, one label for the screening decision and the other label for the editorial comment. Advantageously, the corpus may reflect domain-specific, publisher specific, and/or publication-specific editorial preferences, metrics and/or criteria and is used to train the artificial intelligence module described herein."
        },
        {
            "term": "Variables - manuscripts",
            "definition": "“Variables” refers to measurable or descriptive factors, attributes, inputs, or conditions described in a manuscript that are used in study design, analysis, modeling, or interpretation. Variables may include independent variables, dependent variables, covariates, control variables, demographic factors, experimental conditions, or similar constructs. Variables may be categorical, numerical, ordinal, binary, derived, or otherwise defined."
        },
        {
            "term": "vectorization",
            "definition": "\"Vectorization\" refers to a process of converting words, sentences, or whole bodies of text (i.e. manuscripts), into numerical vectors, typically in a high-dimensional space, so that they can be processed by neural networks. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025)."
        },
        {
            "term": "Vectorized manuscript",
            "definition": "\"Vectorized manuscript\" refers to a representation of a manuscript in a numerical vector format suitable for processing by an artificial intelligence system, neural network, or other machine learning-based computing architecture. A vectorized manuscript may be derived through a transformation pipeline that includes tokenizing the textual content of the manuscript into discrete components (e.g., words, sub words, or tokens), mapping those components to embedding vectors in a high-dimensional space, and assembling the resulting sequence of vectors into a structured format for further computation. The vectorized manuscript may preserve semantic, syntactic, contextual, or positional relationships present in the original manuscript, allowing the system to perform tasks such as classification, generation, summarization, evaluation, or the like. Synonyms for “vectorized manuscript” may include “embedded manuscript,” “numerically encoded manuscript,” “manuscript embeddings,” “manuscript feature vector representation,” or the like. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        },
        {
            "term": "vectorized representations",
            "definition": "\"Vectorized representations\" refers to numerical representations of information—such as text, tokens, token sequences, or other data structures—expressed in the form of vectors in one or more dimensions, typically for use in machine learning models or computational processing. Synonyms include embeddings, feature vectors, encoded vectors, numerical encodings, or the like.\r\n\r\nVectorized representations may be generated through processes such as word embedding, sentence embedding, feature extraction, or deep learning-based encoding and may reflect semantic, syntactic, structural, or contextual characteristics of the input data. Each vector may comprise a plurality of numerical values arranged in a fixed or variable length structure and may be processed using mathematical operations such as dot products, matrix multiplication, or distance functions. The vectorized representations may be input to one or more artificial intelligence systems, including neural networks, transformer models, encoder-decoder pipelines, attention mechanisms, or other computational models, to enable tasks such as classification, similarity analysis, prediction, generation, or the like. Vectorized representations may reside in high-dimensional latent spaces and may evolve during training or inference to capture nuanced relationships between pieces of information. (Defined in conjunction with ChatGPT 4o Version, July 15, 2025.)"
        }
    ]
}