{"id":"W7082991234","doi":"10.5281/zenodo.17183501","title":"La Nuit des Temps Benchmark — LNDT v3.1 | Aeterna Claritas","year":2025,"lang":"fr","type":"report","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Geochemistry and Geologic Mapping","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Benchmark (surveying); Cognition; Certification; Jury","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008408113,0.002222798,0.001249568,0.00206405,0.001598114,0.006433296,0.003317609,0.002169803,0.1879039],"category_scores_gemma":[0.0193252,0.001490863,0.00104788,0.001372887,0.0009246687,0.006948562,0.003699475,0.004477039,0.1530567],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004944726,"about_ca_system_score_gemma":0.007221675,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01521395,"about_ca_topic_score_gemma":0.01464446,"domain_scores_codex":[0.9898276,0.00188701,0.0009485362,0.001009994,0.00531741,0.001009447],"domain_scores_gemma":[0.9892407,0.001446817,0.0003933098,0.003096594,0.004961884,0.0008606719],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0006953824,0.0001576269,0.0005877112,0.0006427937,0.00003727771,0.00008154152,0.0001547452,0.004365454,0.007149526,0.03505775,0.8713841,0.07968601],"study_design_scores_gemma":[0.0001297077,0.0002517261,0.0006685308,0.0003439179,0.00002176773,0.0001244362,0.00007998485,0.01430102,0.01322268,0.008699261,0.96206,0.00009702787],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.00707239,0.002325252,0.1825725,0.004903585,0.004564397,0.003167653,0.03477282,0.306699,0.4539225],"genre_scores_gemma":[0.103666,0.002384676,0.1158092,0.003692719,0.001033295,0.004742542,0.1144905,0.1094227,0.5447585],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.1879039,"threshold_uncertainty_score":0.6286011,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06922302414049236,"score_gpt":0.2659584846529655,"score_spread":0.1967354605124731,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}