{"id":"W39064896","doi":"10.55016/ojs/ajer.v55i1.55272","title":"Peer Observation Reports and Student Evaluations of Teaching: Who Are the Experts?","year":2009,"lang":"en","type":"article","venue":"Alberta Journal of Educational Research","topic":"Evaluation of Teaching Practices","field":"Social Sciences","cited_by":41,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Psychology; Mathematics education; Educational research; Peer evaluation; Higher education; Pedagogy","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02040685,0.00006528582,0.0001451709,0.0002216993,0.001076357,0.0001959123,0.0003627143,0.000053429,0.001111138],"category_scores_gemma":[0.1256309,0.00004748491,0.00005966623,0.0002799225,0.0002446684,0.0008755866,0.00003454346,0.0005259644,0.000006655843],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000184956,"about_ca_system_score_gemma":0.001557262,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001826276,"about_ca_topic_score_gemma":0.001506711,"domain_scores_codex":[0.9938107,0.002119752,0.0005523068,0.0001357336,0.003190117,0.0001914256],"domain_scores_gemma":[0.9768638,0.01843899,0.00073775,0.000208331,0.003625426,0.0001257017],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004513797,0.0008041933,0.5378956,0.000009034016,0.0000876667,8.846988e-7,0.09436115,0.0001635888,0.0004210544,0.2025512,0.1590684,0.004592155],"study_design_scores_gemma":[0.0001725238,0.0001759842,0.7906391,0.0001375283,0.00004393155,0.00002950319,0.03246041,0.0000407105,0.00003493569,0.01651789,0.1596701,0.00007737451],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.542015,0.0005324182,0.000005530396,0.4334985,0.000178778,0.0001597989,2.773286e-7,4.375064e-7,0.02360933],"genre_scores_gemma":[0.9620307,0.0001483672,0.0003302158,0.0005502931,0.0005175362,0.00000787003,0.000002172134,0.000004998755,0.03640782],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4329482,"threshold_uncertainty_score":0.999802,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3669533084308778,"score_gpt":0.6016822682881954,"score_spread":0.2347289598573176,"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."}}