{"id":"W2033420046","doi":"10.1029/2007eo500003","title":"High‐school‐age science assessments","year":2007,"lang":"en","type":"article","venue":"Eos","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"China; Geography; Scale (ratio); Psychology; International comparisons; Demography; Medical education; Political science; Economic growth; Medicine; Sociology; Economics; Cartography","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.001574697,0.0004887044,0.0006960091,0.00298623,0.002552722,0.001426579,0.0006222691,0.0007322211,0.02451646],"category_scores_gemma":[0.004200103,0.0003018237,0.0007057737,0.001621059,0.0003562078,0.0011961,0.001359559,0.0008429536,0.007594016],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006286163,"about_ca_system_score_gemma":0.001970307,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00875342,"about_ca_topic_score_gemma":0.02857533,"domain_scores_codex":[0.9991943,0.0000796706,0.00009134619,0.00009619425,0.0002789165,0.0002595192],"domain_scores_gemma":[0.9955141,0.0003447957,0.0006102724,0.0001865128,0.001876585,0.001467726],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003001207,0.002777245,0.9137745,0.0001884631,0.00005864117,0.0004223702,0.003276966,0.0002253015,0.001149679,0.0007620514,0.01609948,0.06096515],"study_design_scores_gemma":[0.00002770463,0.0004756706,0.9799412,0.00004451906,0.0000193706,0.0001233606,0.001308794,0.0001207266,0.0009457258,0.000255634,0.01671658,0.00002066767],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.95385,0.0002927467,0.0003270496,0.0003023712,0.0001043001,0.0004588548,0.005671482,0.0001284836,0.03886474],"genre_scores_gemma":[0.9534233,0.0003560904,0.001417291,0.0002465,0.00004681918,0.00043847,0.005137787,0.0000161992,0.03891756],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9984253,"threshold_uncertainty_score":0.08201575,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1379297661429994,"score_gpt":0.5092086912751245,"score_spread":0.3712789251321251,"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."}}