1 000 travaux · Opus 4.8 · GPT-5.6 (high) · Grok 4.5 · une seule grille verrouillée
La frontière du domaine n'est pas une ligne. C'est une région.
Trois modèles de pointe ont trié les mêmes 1 000 travaux, tirés de la vraie base avec des probabilités de sélection connues, selon la même grille verrouillée sur sa charge utile complète de huit champs. Parmi les 51 travaux qu'au moins un modèle a qualifiés de métarecherche, seulement 19 (37,3 %) l'ont été par les trois, et 24 (47,1 %) reposent sur l'avis d'un seul modèle.
Deux trieurs peuvent s'accorder sur un taux tout en repérant des travaux presque entièrement différents. À un taux de base d'environ 1 %, les rejets évidents offrent 98 % d'accord gratuitement ; c'est pourquoi une statistique d'accord calculée sur tout l'échantillon ne dit rien de la frontière, et pourquoi le livrable est le tableau ci-dessous, et non un pourcentage.
Les trois modèles ne sont pas interchangeables
Combien des 1 000 travaux chaque modèle a qualifiés de métarecherche (niveau T1 ou T2), sur des données identiques. T3 est adjacent et ne compte pas comme dans le champ ; c'est pourquoi le compte d'un modèle ne peut jamais dépasser les 51 travaux du dossier.
Changez le modèle que vous appelez « le trieur » et la taille du domaine bouge. Cet écart, et non l'intervalle de confiance binomial sur les étiquettes d'un seul modèle, est l'incertitude honnête sur la taille de la métarecherche canadienne.
Le dossier des désaccords
Les trois verdicts, confiances et motifs côte à côte. Ce sont les travaux sur lesquels les critères d'inclusion doivent être rédigés, parce que ce sont ceux sur lesquels des trieurs raisonnables, devant la même grille et les mêmes preuves, divergent.
| Travail | Strate | n_in | Opus 4.8 | GPT-5.6 | Grok 4.5 |
|---|---|---|---|---|---|
| L’éducation thérapeutique du patient pour les clientèles psychiatriques présentant un surpoids : l’exemple du programme « Wellness » développé au Québec 2016 · article · fr · L Encéphale | french w=1554 | 0/3 | OUTmedium empirical No abstract; a therapeutic patient-education program for psychiatric patients with overweight, i.e. clinical intervention. | OUTlow other The abstract is missing and the title indicates a clinical patient-education program rather than research about research. | OUTmedium empirical Clinical therapeutic-education program for psychiatric patients with overweight; empty abstract but clinical object is clear. |
| Stress scolaire, soutien social et burnout à l’adolescence : quelles relations? 2016 · article · fr · Éducation et francophonie | french w=1554 | 0/3 | OUThigh empirical Study of school stress, social support and burnout in adolescents. | OUThigh empirical It studies stress, social support, and school burnout among adolescents. | OUThigh empirical Adolescent school stress, social support, and burnout is educational psychology. |
| Sakahan: International Indigenous Art [Book Review] 2014 · article · en · Artlink | about_only w=3321 | 0/3 | OUThigh other Review of an international Indigenous art exhibition; an arts review. | OUThigh editorial/commentary It reviews an Indigenous art exhibition, not research practice. | OUThigh editorial/commentary Art exhibition book review; cultural criticism, not metaresearch. |
| Waterfront News 1(2) 2007 · article · en · York University Digital Library (York University) | about_only w=3321 | 0/3 | T3low other Newsletter issue reporting the current activities of a Toronto waterfront research project; contextual research-project news rather than a study of research. | T3high infrastructure/announcement This newsletter reports activities of a research project and is contextual rather than analytic. | T3high infrastructure/announcement Project newsletter announcing activities of a Toronto waterfront research project. |
| Effects of Apomorphine on Subthalamic Nucleus and Globus Pallidus Internus Neurons in Patients With Parkinson's Disease 2001 · article · en · Journal of Neurophysiology | aff_core w=5595 | 0/3 | OUThigh empirical Neurophysiology of apomorphine effects on basal ganglia neurons in Parkinson's disease. | OUThigh empirical This is a clinical neuroscience study of apomorphine effects in Parkinson's disease. | OUThigh empirical Clinical neurophysiology study of apomorphine effects on basal ganglia neurons. |
| Strengthening the Internet for Global, Ubiquitous and Secure Commercial Use: Perspectives, Lessons, Issues and Challenges 2016 · article · en · SSRN Electronic Journal | aff_core w=5595 | 0/3 | OUThigh conceptual Policy analysis of internet governance and trust for commercial use; the governed system is the internet, not research. | OUThigh conceptual The article discusses Internet governance and security for commercial use, not research infrastructure or research practice. | OUThigh policy Telecom/Internet commercial trust and legal infrastructure for digital markets; ICT policy, not metaresearch. |
| A MILP model for joint product family and supply chain design 2011 · preprint · en · HAL (Le Centre pour la Communication Scientifique Directe) | aff_core w=5595 | 0/3 | OUThigh empirical MILP optimization model for joint product family and supply chain design; operations research on a design problem, not on research. | OUThigh conceptual The work proposes a supply-chain design model and does not study research itself. | OUTmedium empirical Title is joint product-family and supply-chain optimization; domain operations research with thin abstract. |
| 10.1146/annurev-pathol-011110-130211 2000 · article · en · Time to knit | venue_new w=2684 | 0/3 | OUTlow other Malformed record whose title is a bare DOI with no usable content; nothing indicates metaresearch. | OUTlow other The record contains only a DOI-like title and no abstract, so its object cannot be established. | OUTlow other Garbled DOI-like title with no abstract; cannot support metaresearch inclusion. |
| The Time and Sound of the Nautical Border 2019 · book-chapter · en · University of British Columbia Press eBooks | about_only w=3321 | 0/3 | OUThigh conceptual Book chapter in a postcolonial history collection on the Komagata Maru; the object is migration history and Canadian multiculturalism, not the research system. | OUThigh conceptual The chapter analyzes the historical and colonial significance of the Komagata Maru incident rather than research. | OUThigh conceptual Historical/cultural analysis of the Komagata Maru and migration; Canadian history, not the research system. |
| Insulin Detemir Decreases Gain in Body Weight and Fat Mass in ZDF Rats when Compared with both NPH and Insulin Glargine 2008 · article · en · Canadian Journal of Diabetes | venue_new w=2684 | 0/3 | OUThigh empirical Preclinical comparison of insulin formulations in rats; abstract absent but the title is unambiguous. | OUThigh empirical This is a biomedical animal study of insulin effects on weight and fat mass. | OUThigh empirical Title is an animal insulin/weight pharmacology comparison; domain biomedical research. |
| Patients' narratives of open-heart surgery : emplotting the technological. 2009 · dissertation · en · Library and Archives Canada (Government of Canada) | venue_new w=2684 | 0/3 | OUTmedium empirical Narrative study of patients' experience of technology in heart surgery; studies clinical care, not scientific practice, so it falls outside the STS-of-research sense of T2. | OUThigh empirical The work studies patients' narratives and experiences of open-heart surgery technology. | OUThigh empirical Narrative study of patients’ experience of technology in open-heart surgery is clinical/medical sociology, not study of scientific research practice. |
| <i>Sympathy, Sensibility and the Literature of Feeling in the Eighteenth Century</i> by Ildiko Csengei 2016 · article · en · Eighteenth-Century Fiction | venue_new w=2684 | 0/3 | OUThigh other Book review of an eighteenth-century literary studies monograph; the object is literary sensibility, not research practice. | OUThigh editorial/commentary This is a literary book review unrelated to research practice. | OUThigh editorial/commentary Book review of eighteenth-century literature of feeling, not research practice. |
| High Resolution Adaptively Steered Horizon Tracker 2012 · article · en | aff_core w=5595 | 0/3 | OUThigh empirical Signal processing method for automatic horizon tracking in 3D seismic interpretation; the object is a geophysical interpretation tool. | OUThigh empirical This work develops an automated seismic horizon tracker for geophysical interpretation, not metaresearch. | OUThigh empirical Seismic interpretation algorithm for automatic horizon tracking; domain geophysics tool. |
| Cellular and Molecular Cascades during Liver Regeneration 2016 · article · en · Surgical Research - Open Journal | aff_core w=5595 | 0/3 | OUThigh empirical Review of cellular and molecular cascades during liver regeneration; a biomedical domain review. | OUThigh other It reviews biological mechanisms of liver regeneration, not research practice. | OUThigh empirical Review of cellular/molecular liver regeneration; biomedical biology. |
| Queering Law: Not by <i>Vriend</i> 2002 · article · fr · Canadian Journal of Law and Society / Revue Canadienne Droit et Société | french w=1554 | 0/3 | OUThigh conceptual Queer-theory legal analysis of a Supreme Court decision on sexual orientation; not about research. | OUThigh conceptual The article analyzes queer theory and Canadian law rather than research practice. | OUThigh conceptual Queer legal theory analysis of a Canadian rights case; object is law and identity, not research practice. |
| Dynamic Train-Bridge Interaction in Monorail Sao Paulo Metro Line 2 2013 · article · en · Report | about_only w=3321 | 0/3 | OUThigh empirical Numerical simulation of train-bridge dynamic interaction; civil engineering. | OUThigh empirical It models train-bridge dynamics for engineering design rather than studying research. | OUThigh empirical Dynamic train–bridge interaction modelling for a metro monorail; civil engineering. |
| Worte zum Bild "Rehkitz" 2013 · article · de · ti< | venue_new w=2684 | 0/3 | OUThigh other A poem; not research at all. | OUThigh other This is a poem and does not study research. | OUThigh other Literary poem with no research object. |
| Michel O’neill, L’épopée des Petits frères de la Croix. Histoire d’une nouvelle communauté monastique québécoise dans l’Église catholique d’aujourd’hui, Québec, Les Presses de l’Université Laval, 2014, 232 p. 2015 · article · fr · Recherches sociographiques | french w=1554 | 0/3 | OUThigh other French-language book review of a history of a Quebec monastic community. | OUThigh editorial/commentary It is a book review about a Québec monastic community, not research practice. | OUThigh editorial/commentary Book notice on a Quebec monastic community history; not about research practice. |
| The Murals of Moose Jaw: Commodification or Articulation of the Past? 2000 · article · en · Historical geography | about_only w=3321 | 0/3 | OUThigh empirical Cultural geography study of the murals of Moose Jaw as commodification of the past. | OUThigh empirical It studies historical murals and local economic development rather than research practice. | OUThigh empirical Historical geography of tourism murals and local economic revitalization, not metaresearch. |
| Phthalocyanine-Based Organic Thin-Film Transistors: A Review of Recent Advances 2015 · review · en · ACS Applied Materials & Interfaces | aff_core w=5595 | 0/3 | OUThigh empirical Materials-science review of phthalocyanine organic thin-film transistors; a domain review. | OUThigh other It reviews organic-electronics materials and devices, not research practice. | OUThigh empirical Materials review of phthalocyanine organic transistors; domain electronics science. |
| Preliminary Study on Seed Propragtion of Louisiana Iris 2010 · article · en · Seed | venue_new w=2684 | 0/3 | OUThigh empirical No abstract; horticultural study of iris seed propagation. | OUTlow empirical The title indicates a study of Louisiana iris seed propagation, not research practice. | OUTmedium empirical Horticultural seed propagation of Louisiana iris; plant cultivation (no abstract). |
| Effects of fructus forsythiae extracts on scavenging oxygen free radical and aliphatic free radical 2008 · article · en | aff_core w=5595 | 0/3 | OUThigh empirical Phytochemistry study of free radical scavenging by Forsythia fruit extracts. | OUThigh empirical It evaluates antioxidant activity of a plant extract, not research itself. | OUThigh empirical Phytochemical antioxidant assay of plant extracts; domain chemistry/biology. |
| Research on the Development Idea for the Key Seed Enterprises in Henan Province 2013 · article · en · Seed | venue_new w=2684 | 0/3 | OUTmedium other Development strategy for seed enterprises in Henan; an agribusiness-development note with no abstract, and 'research' here is corporate strategy. | OUTlow conceptual The title indicates a study of seed enterprises in Henan Province, but the abstract is missing. | OUTlow other Title concerns development ideas for seed enterprises in Henan; domain agribusiness, abstract missing. |
| [Etiology of intestinal occlusion]. 2003 · article · en · PubMed | about_only w=3321 | 0/3 | OUThigh empirical Retrospective cohort on the etiology of intestinal obstruction. | OUThigh empirical It studies the causes of intestinal obstruction in clinical patients. | OUThigh empirical Retrospective clinical series on causes of intestinal obstruction; Canada appears only in comparative literature. |
| Main Characteristics and High Yield Cultivated Technology of New Sweet Potatoes Variety Longshu No.9 2011 · article · en · Seed | venue_new w=2684 | 0/3 | OUThigh other Description of a new sweet potato cultivar and its cultivation; agronomy. | OUThigh other The title concerns sweet-potato cultivation technology rather than research practice. | OUThigh other Agricultural variety description of sweet potato Longshu No.9; crop breeding announcement-style domain work. |
| Effects of 2,4-D Seed Soaking Treatment on Leek Seed Germination and Seedling Early Growth 2013 · article · en · Seed | venue_new w=2684 | 0/3 | OUThigh empirical Horticultural experiment on 2,4-D seed soaking and leek germination; the object is plant growth. | OUThigh empirical The experiment studies herbicide effects on leek germination rather than research practice. | OUThigh empirical Seed science experiment on 2,4-D soaking and leek germination; plant physiology. |
| The Argument on The General Education and Special Education in Social Work Field of China 2007 · article · en | aff_core w=5595 | 0/3 | OUTmedium conceptual Argument about general versus specialized professional education in Chinese social work; concerns vocational training, not the research workforce. | OUThigh conceptual The work debates general and specialized social-work education rather than research itself. | OUThigh conceptual Debate on general vs specialized social-work professional education in China, not scientific research workforce. |
| Clustering-based redshift estimation: application to VIPERS/CFHTLS 2016 · article · en · Monthly Notices of the Royal Astronomical Society | about_only w=3321 | 0/3 | OUThigh empirical Clustering-based redshift estimation applied to galaxy surveys; an astronomy method used to answer an astronomy question. | OUThigh empirical The work tests a redshift-estimation technique for galaxy surveys rather than studying research practice. | OUThigh empirical Astronomy method for clustering-based redshift estimation; domain astrophysics technique. |
| Generating Examples for Knowledge Abstraction in MDE: a Multi-Objective Framework. 2015 · article · en | aff_core w=5595 | 0/3 | OUThigh empirical Model-driven software engineering method for generating model examples; a software technique, not a study of research. | OUThigh empirical The paper develops an automated model-generation framework for software engineering, not a study of research itself. | OUThigh empirical Software-engineering method for generating MDE model examples; SE tooling, not study of how research is done. |
| New molecular assays for cancer diagnosis and targeted therapy. 2008 · article · en · PubMed | aff_core w=5595 | 0/3 | OUThigh other Review of molecular assays for cancer diagnosis and targeted therapy; the object is diagnostic technology and its clinical regulation. | OUThigh conceptual The review concerns cancer diagnostic assays and treatments, not research methodology. | OUThigh other Industry-oriented review of molecular diagnostic cancer assays and personalized medicine testing. |
| Entecavir plus tenofovir combination therapy for chronic hepatitis B in patients with previous nucleos(t)ide treatment failure 2016 · article · en · Hepatology International | aff_core w=5595 | 0/3 | OUThigh empirical Clinical study of entecavir plus tenofovir in chronic hepatitis B after treatment failure; a therapeutic efficacy question. | OUThigh empirical This clinical study concerns combination therapy for chronic hepatitis B, not research practice. | OUTlow empirical Clinical hepatology combination therapy for chronic hepatitis B; domain medicine, abstract missing. |
| Gene evolution and gene expression after whole genome duplication in fish: the PhyloFish database 2016 · article · en · BMC Genomics | fund_new w=1679 | 0/3 | OUTmedium empirical Transcriptome analyses plus release of the PhyloFish database; a domain genomics resource, not infrastructure of the research ecosystem in the metaresearch sense. | T3high infrastructure/announcement This describes a genomics database resource and is an infrastructure announcement rather than an analytic study. | OUThigh empirical Domain genomics resource (PhyloFish) for fish gene evolution; uses a database tool rather than studying research infrastructure. |
| Application of Non‐Linear Automatic Optimization Techniques for Calibration of HSPF 2007 · article · en · Water Environment Research | aff_core w=5595 | 0/3 | OUTmedium empirical Comparison of optimization algorithms for calibrating a watershed model; a domain modeling technique, not a study of research practice. | OUThigh empirical The work compares techniques for calibrating a hydrological model rather than studying research methods generally. | OUThigh empirical Automatic parameter calibration of a watershed hydrology model; domain environmental modelling, not metaresearch methods. |
| Dual ligand/receptor interactions activate urothelial defenses against uropathogenic E. coli 2015 · article · en · Scientific Reports | fund_new w=1679 | 0/3 | OUThigh empirical Molecular immunology of urothelial defense against uropathogenic E. coli. | OUThigh empirical This is a biomedical study of urothelial defenses against infection. | OUThigh empirical Host–pathogen immunology study of urothelial defenses against UPEC. |
| La idealidad en Husserl: la lógica y las esencias necesarias 2012 · article · es · Alpha Omegan | venue_new w=2684 | 0/3 | OUThigh conceptual Philosophical analysis of Husserl's doctrine of ideality; not about contemporary research practice. | OUThigh conceptual This is a philosophical analysis of Husserl's theory of ideality, not research itself. | OUThigh conceptual Husserl phenomenology of ideality; pure philosophy, not contemporary research practice. |
| The Preference for Internet-Based Psychological Interventions by Individuals Without Past or Current Use of Mental Health Treatment Delivered Online: A Survey Study With Mixed-Methods Analysis 2016 · article · en · JMIR Mental Health | venue_new w=2684 | 0/3 | OUThigh empirical Survey of preferences for internet-based psychological treatment; a health-services question. | OUThigh empirical This survey studies preferences for mental-health interventions rather than research itself. | OUThigh empirical Survey of patient preference for internet-based psychological treatment versus face-to-face care. |
| The Association between Medical Education Accreditation and the Examination Performance of Internationally Educated Physicians Seeking Certification in the United States. 2012 · book-chapter · en · TUScholarShare (Temple University) | about_only w=3321 | 0/3 | OUTmedium empirical Study of medical-school accreditation and exam performance of international medical graduates; the object is clinical training and its quality assurance, not the research system or research workforce. | OUThigh empirical This studies medical education accreditation and examination performance, not research itself. | OUThigh empirical Studies medical-school accreditation and USMLE performance of clinicians-in-training, not research methods, evaluation, or the research workforce. |
| Generation of Lactococcus lactis capable of coexpressing epidermal growth factor and trefoil factor to enhance in vitro wound healing 2015 · article · en · Applied Microbiology and Biotechnology | aff_core w=5595 | 0/3 | OUThigh empirical Microbiology engineering of Lactococcus lactis for wound healing; abstract absent but the title is unambiguous. | OUThigh empirical It reports a microbiology and wound-healing experiment, not research practice. | OUThigh empirical Title indicates applied microbiology wound-healing work; empty abstract but domain object is unambiguous. |
| [The right not to know, in the German legislation (Part I)]. 2005 · article · en · PubMed | aff_core w=5595 | 0/3 | OUTmedium conceptual Legal analysis of the right not to know in German legislation; medical law and bioethics, not research practice. | OUThigh conceptual It analyzes a legal and ethical right in German legislation. | OUThigh conceptual Legal-ethical discussion of the right not to know in German law, not research practice. |
| Bullying Experiences Among First Nations Youth: Identifying Effects on Mental Health and Potential Protective Factors 2016 · article · en · Scholarship@Western (Western University) | about_only w=3321 | 0/3 | OUThigh empirical Primary study of bullying experiences and mental health among First Nations youth; the object is bullying, not research practice. | OUThigh empirical The study examines bullying and mental health among First Nations, Metis and Inuit youth, not research itself. | OUThigh empirical Social/developmental study of bullying among Indigenous youth, not the research system. |
| Heavy metal capture by autochthonous yeasts from a volcanic influenced environment of Patagonia 2016 · article · en · Journal of Basic Microbiology | fund_new w=1679 | 0/3 | OUThigh empirical Microbiology study of heavy metal capture by yeasts; the object is bioremediation. | OUThigh empirical The study investigates heavy-metal tolerance and capture by yeasts. | OUThigh empirical Microbiology of heavy-metal capture by yeasts. |
| Reliability of the SWAG—The Standardized Way to Assess Grafts Method for Alveolar Bone Grafting in Patients with Cleft Lip and Palate 2016 · article · en · The Cleft Palate-Craniofacial Journal | aff_core w=5595 | 0/3 | OUTmedium empirical Intra- and inter-rater reliability of a radiographic scale for grading bone grafts; clinical instrument reliability in the measurement sense, a polysemy trap rather than metaresearch. | OUThigh empirical This study tests the reliability of a clinical graft-assessment instrument rather than research practice. | OUThigh empirical Interrater reliability of a clinical graft-assessment scale; clinical measurement validation, polysemy trap not research reproducibility. |
| Root gerunds in Old Romanian 2016 · book-chapter · en · Romance languages and linguistic theory | aff_core w=5595 | 0/3 | OUThigh empirical Theoretical syntax analysis of root gerunds in Old Romanian; the object is language, not research. | OUThigh conceptual The work studies Old Romanian syntax rather than research practice. | OUThigh empirical Historical linguistics of Old Romanian syntax, not research as object. |
| The Chloroplast Protease AMOS1/EGY1 Affects Phosphate Homeostasis under Phosphate Stress 2016 · article · en · PLANT PHYSIOLOGY | aff_core w=5595 | 0/3 | OUThigh empirical Plant molecular biology of phosphate homeostasis in Arabidopsis. | OUThigh empirical The study investigates phosphate homeostasis in Arabidopsis plants. | OUThigh empirical Plant molecular physiology of a chloroplast protease under phosphate stress; domain biology. |
| Language, Literacy and Communication in Nigeria: Minority Languages and National Development 2002 · article · en · Papers from the Annual Meetings of the Atlantic Provinces Linguistic Association (PAMAPLA) | aff_core w=5595 | 0/3 | OUThigh conceptual Essay on minority languages, literacy policy and national development in Nigeria. | OUThigh conceptual This paper discusses language, literacy, and development in Nigeria, not research. | OUThigh conceptual Language and literacy policy in Nigeria; not research or scholarly communication as object. |
| Transcatheter or surgical aortic valve replacement for patients with severe, symptomatic, aortic stenosis at low to intermediate surgical risk: a clinical practice guideline 2016 · article · en · BMJ | aff_core w=5595 | 0/3 | OUThigh policy Clinical practice guideline on aortic valve replacement; recommendations answering a clinical question. | OUThigh policy This clinical practice guideline answers a treatment question using systematic reviews rather than studying synthesis methodology. | OUThigh policy Clinical practice guideline on aortic valve replacement drawing on systematic reviews to answer a care question. |
| Changing the Learning Curve in Novice Laparoscopists: Incorporating Direct Visualization into the Simulation Training Program 2016 · article · en · Journal of surgical education | aff_core w=5595 | 0/3 | OUTmedium empirical Surgical simulation training study; clinical skills education, not research training or the research workforce. | OUThigh empirical The study concerns surgical simulation training for novice laparoscopists, not research itself. | OUTmedium empirical Surgical skills simulation training for novice laparoscopists is clinical education, not study of the research workforce or research methods. |
| The Role of Environmental Stewardship Groups in the Grand River Watershed, Ontario 2014 · dissertation · en · UWSpace (University of Waterloo) | about_only w=3321 | 0/3 | OUThigh empirical Thesis on environmental stewardship volunteer groups in a watershed. | OUThigh empirical This studies environmental stewardship groups and watershed effects, not research itself. | OUThigh empirical Thesis on environmental stewardship groups in a watershed; environmental management domain. |
| The membrane scaffold SLP2 anchors a proteolytic hub in mitochondria containing PARL and the i‐AAA protease YME1L 2016 · article · en · EMBO Reports | aff_core w=5595 | 0/3 | OUThigh empirical Mitochondrial membrane scaffold and protease complex; cell biology. | OUThigh empirical This studies a mitochondrial proteolytic complex, not research practice. | OUTmedium empirical Cell/molecular biology of mitochondrial protease scaffolding; title clear domain science, abstract missing. |
| Visualization of dominant stress-transfer mechanisms in experimental debris flows of different particle-size distribution 2016 · article · en · Canadian Geotechnical Journal | venue_new w=2684 | 0/3 | OUThigh empirical Physical modeling of stress transfer in experimental debris flows. | OUThigh empirical This experimentally studies stress transfer mechanisms in debris flows, not research practice. | OUThigh empirical Physical modelling of debris-flow stress transfer; geotechnical domain science. |
Ce que le tri a réellement trouvé
Trois modèles de pointe (Opus 4.8, GPT-5.6 high, Grok 4.5) ont trié les mêmes 1 000 travaux de la VRAIE base de 4,3 M, sur la charge utile COMPLÈTE de huit champs de la grille, avec des lots randomisés consignés au manifeste et des étiquettes écrites par le dispositif : chaque défaut relevé par D1, D2, D11 et le constat 16, réparé. Les taux de base pondérés par le plan de sondage s'étendent de 1,89 % à 3,54 % (1,9 fois). Mais ce sont les ensembles qui font le constat, comme le constat 16 le prédisait : des 51 travaux qu'AU MOINS UN modèle a qualifiés de métarecherche, seulement 19 (37 %) l'ont été par LES TROIS, et 24 (47 %) reposent sur l'avis d'un seul modèle. LA FRONTIÈRE DU DOMAINE N'EST PAS UNE LIGNE QUE LES MODÈLES PARTAGENT ; C'EST UNE RÉGION QUE CHACUN DÉCOUPE À SA FAÇON. GPT-5.6 a de plus enfreint le schéma de sortie verrouillé sur 18 des 1 000 notices, en écrivant des valeurs de genre dans le champ du niveau, ce que le validateur de manifeste a détecté. Le livrable n'est pas le taux de base : c'est le dossier des désaccords, les 51 travaux qui marquent la frontière empirique du domaine et sur lesquels les critères d'inclusion doivent réellement être rédigés.
Calculé le 2026-07-12T19:52:14Z · les 22 constats · ce dossier en JSON