{"id":"W2990689782","doi":"10.5753/cbie.sbie.2019.199","title":"Avaliação de Juízes: Um Modelo Estatístico para Perfilação de Avaliadores","year":2019,"lang":"pt","type":"article","venue":"Anais do XXX Simpósio Brasileiro de Informática na Educação (SBIE 2019)","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Canadian Bureau for International Education","keywords":"Concordance; Competence (human resources); Statistics; Psychology; Mathematics; Computer science; Social psychology; Medicine","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"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":[0.01626832,0.001311528,0.001205871,0.003918572,0.00103666,0.005279914,0.002229894,0.001867468,0.008080332],"category_scores_gemma":[0.0760897,0.0009937666,0.002526972,0.004045921,0.001451535,0.004937514,0.002184267,0.002392265,0.001310378],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002829684,"about_ca_system_score_gemma":0.004812497,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0136046,"about_ca_topic_score_gemma":0.01425048,"domain_scores_codex":[0.9890315,0.006353465,0.0005722745,0.001682032,0.001850965,0.0005098101],"domain_scores_gemma":[0.9328867,0.0529807,0.004777761,0.00321698,0.005355135,0.0007827089],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001355944,0.001267379,0.3099867,0.002825675,0.001332239,0.0004551568,0.0164705,0.2324856,0.004457063,0.07785311,0.007682885,0.3438278],"study_design_scores_gemma":[0.0002006083,0.001741957,0.1128009,0.0007950452,0.0008369058,0.0004580849,0.004925185,0.7945675,0.002513179,0.06258772,0.01831314,0.0002596592],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3221841,0.001419298,0.6478744,0.003664454,0.0002396985,0.00182715,0.002207239,0.001756753,0.01882691],"genre_scores_gemma":[0.8553364,0.0005356803,0.1373897,0.0001688655,0.00004862695,0.001319123,0.0005782711,0.0001319329,0.004491388],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01626832,"threshold_uncertainty_score":0.08603609,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0738775683116693,"score_gpt":0.35990389654442,"score_spread":0.2860263282327506,"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."}}