{"id":"W2995076146","doi":"10.5539/ies.v13n1p1","title":"Not All Finns Think Alike: Varying Views of Assessment in Finland","year":2019,"lang":"en","type":"article","venue":"International Education Studies","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Narrative; Psychology; Socioeconomic status; Pedagogy; Standardized test; Narrative inquiry; Teacher education; Medical education; Educational assessment; Mathematics education; Sociology; Medicine","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":[],"consensus_categories":[],"category_scores_codex":[0.0005854316,0.00008009265,0.0001709099,0.0001446492,0.00006483685,0.00003669032,0.000254566,0.00003511638,0.0006650275],"category_scores_gemma":[0.0001939927,0.0000746618,0.00005112224,0.0001491994,0.00006753134,0.0002329821,0.00007550196,0.00009258381,0.00009533695],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002401395,"about_ca_system_score_gemma":0.0003204507,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005974701,"about_ca_topic_score_gemma":0.0003875446,"domain_scores_codex":[0.9988323,0.00007945791,0.0002933152,0.0001674215,0.0004996462,0.0001278641],"domain_scores_gemma":[0.9989434,0.0002580778,0.0001702379,0.00009437627,0.0005097721,0.00002409146],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001721671,0.0005363693,0.8628122,0.00002874808,0.0002578527,5.235103e-7,0.06535894,0.00004074048,0.0004999459,0.05108569,0.01541866,0.003943055],"study_design_scores_gemma":[0.0007682416,0.00005166616,0.604663,0.0001841362,0.0000245361,4.105237e-7,0.06881531,0.00005741878,0.0001219288,0.004411962,0.3206624,0.0002390402],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7956247,0.0005921924,0.00002231907,0.009968637,0.005477742,0.000293278,0.000004957482,0.00001436882,0.1880019],"genre_scores_gemma":[0.9797078,0.0007814856,0.001233805,0.0006209346,0.0002066055,0.00004007546,0.00001143853,0.000004957631,0.01739286],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3052438,"threshold_uncertainty_score":0.7281585,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1321783318689788,"score_gpt":0.501713879522738,"score_spread":0.3695355476537593,"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."}}