{"id":"W2795852592","doi":"10.1111/emip.12198","title":"A Review of Recent Research on Individual‐Level Score Reports","year":2018,"lang":"en","type":"review","venue":"Educational Measurement Issues and Practice","topic":"Educational and Psychological Assessments","field":"Psychology","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Context (archaeology); Test (biology); Computer science; Focus (optics); Knowledge management; Psychology; Data science","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01566091,0.0003925577,0.0009974786,0.0003027703,0.000211398,0.00006467659,0.0004097661,0.0002763655,0.01204313],"category_scores_gemma":[0.007893226,0.0002938593,0.0001852352,0.0008379045,0.0002189225,0.0001383613,0.00009846828,0.0008362135,0.0007113995],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002051836,"about_ca_system_score_gemma":0.0013002,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001067442,"about_ca_topic_score_gemma":0.000002543984,"domain_scores_codex":[0.9918364,0.003377513,0.001277481,0.0009589893,0.002159024,0.0003906247],"domain_scores_gemma":[0.9909995,0.003730173,0.001322876,0.0008715204,0.002866091,0.0002098265],"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":[0.0000367291,0.002057377,0.00001568505,0.02171743,0.0005696143,0.000006475711,0.0002928548,3.620432e-9,7.690469e-8,0.006812384,0.6030199,0.3654715],"study_design_scores_gemma":[0.00006889992,0.0003483306,0.0005875271,0.07285575,0.0005657261,0.0002989682,0.0001964128,1.27923e-9,1.716711e-7,0.001785569,0.9230408,0.0002518554],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000001544391,0.8878733,9.895169e-7,0.01559044,0.002726556,0.001321125,0.00007251647,0.0000104841,0.09240305],"genre_scores_gemma":[0.000004775834,0.9897828,0.0009371408,0.002479854,0.001724006,0.0008626519,0.0004010241,0.00003893572,0.003768816],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.3652196,"threshold_uncertainty_score":0.9999514,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8619976910036018,"score_gpt":0.6550214592649924,"score_spread":0.2069762317386094,"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."}}