{"id":"W7105762787","doi":"10.5281/zenodo.17605286","title":"RHEA–UCM Thermodynamic Benchmark White Paper & Curriculum","year":2025,"lang":"","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Advanced Thermodynamics and Statistical Mechanics","field":"Physics and Astronomy","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Teck (Canada)","funders":"","keywords":"License; Benchmarking; Benchmark (surveying); Certification; Reciprocity (cultural anthropology); Class (philosophy); Database transaction; Entropy (arrow of time); White paper","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0005496474,0.0004574239,0.0004398777,0.0002686729,0.003778087,0.001214432,0.001299087,0.000152927,0.04332089],"category_scores_gemma":[0.0001878749,0.0004912391,0.0002023308,0.001075181,0.0002677319,0.000370507,0.001945239,0.0008121109,0.004673752],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003072703,"about_ca_system_score_gemma":0.00001867975,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003321426,"about_ca_topic_score_gemma":6.592623e-7,"domain_scores_codex":[0.9965231,0.0005013309,0.0006477039,0.0009598825,0.0004423011,0.000925685],"domain_scores_gemma":[0.9976204,0.0000607095,0.0002342409,0.0009535085,0.0008064982,0.0003246329],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00007052744,0.0005590434,0.000007639057,0.0001179404,0.0002344133,0.00001045685,0.000460619,0.0002855085,0.002235261,0.6629312,0.016907,0.3161804],"study_design_scores_gemma":[0.001118711,0.000294058,0.0004306526,0.0002405689,0.0001232783,0.00001182706,0.001165162,0.09147852,0.00004149692,0.06128113,0.8431977,0.0006168764],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.008451802,0.0003292456,0.5490569,0.001378135,0.0007061538,0.001090517,0.001978289,0.0004115956,0.4365973],"genre_scores_gemma":[0.9893581,0.0001778822,0.0009073018,0.0002642659,0.0002405832,1.972703e-7,0.00318108,0.001600654,0.004269951],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9809063,"threshold_uncertainty_score":0.9998224,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.008952153758300918,"score_gpt":0.2411109354377935,"score_spread":0.2321587816794926,"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."}}