{"id":"W7105738225","doi":"10.5281/zenodo.17605287","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01445117,0.001161033,0.001117135,0.0027384,0.002304011,0.007791168,0.005005442,0.003064639,0.113386],"category_scores_gemma":[0.02851374,0.0007515509,0.001098461,0.00254295,0.001775322,0.005712111,0.004933198,0.003666501,0.08453162],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004700703,"about_ca_system_score_gemma":0.01016426,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005402658,"about_ca_topic_score_gemma":0.00494084,"domain_scores_codex":[0.9911742,0.002185116,0.0002928188,0.0007862483,0.004851594,0.0007099735],"domain_scores_gemma":[0.9864033,0.001415391,0.0004885013,0.003838667,0.006699902,0.001154166],"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":[0.0001429212,0.0002114256,0.0004588473,0.0003078255,0.00002037852,0.00005539209,0.00009010956,0.009157521,0.00105505,0.1721552,0.7073767,0.1089687],"study_design_scores_gemma":[0.00003763033,0.0001293611,0.0005259567,0.0004262517,0.000008024087,0.00004658403,0.00008113335,0.01022383,0.003475757,0.04779226,0.9372047,0.00004850195],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"methods","genre_scores_codex":[0.008135869,0.003722807,0.1163624,0.02799533,0.008204963,0.001727125,0.01947468,0.01576956,0.7986072],"genre_scores_gemma":[0.1115139,0.006743528,0.1834475,0.009435816,0.003913557,0.003943857,0.05102039,0.02121285,0.6087686],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.113386,"threshold_uncertainty_score":0.379314,"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."}}