{"id":"W4316589304","doi":"10.7202/1095480ar","title":"(In)Stability of Test Scores","year":2023,"lang":"en","type":"article","venue":"Canadian Journal of Educational Administration and Policy","topic":"Pasture and Agricultural Systems","field":"Agricultural and Biological Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Test (biology); Statistic; Standardized test; Cohort; Psychological intervention; Achievement test; Psychology; Scale (ratio); Test score; Mathematics education; Statistics; Mathematics; Geography","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0001469474,0.00004556213,0.00008805815,0.0000499336,0.00005408776,0.00002382257,0.00007128738,0.00003202444,0.0002564767],"category_scores_gemma":[0.0003303539,0.00001816991,0.00003252739,0.0004329387,0.00004414483,0.0001051972,0.000002497691,0.00004811876,0.000006802587],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002837014,"about_ca_system_score_gemma":0.0005006673,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006156106,"about_ca_topic_score_gemma":0.08167381,"domain_scores_codex":[0.9995141,0.0000233652,0.0002340636,0.00005375794,0.00007205696,0.0001026802],"domain_scores_gemma":[0.9993174,0.0002528532,0.0001119338,0.00001393157,0.00009748738,0.0002064051],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001121631,0.0000712424,0.8971248,0.00002335213,0.000008842547,0.00000747169,0.000663672,0.000001881451,0.06552213,0.01597686,0.01723826,0.003350316],"study_design_scores_gemma":[0.00004273983,0.0001229469,0.9840418,0.00002151217,0.000001742354,0.00004609434,0.0003119718,0.000001183499,0.0008638848,0.001666611,0.01283988,0.00003967043],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9503175,0.00008922189,1.807645e-8,0.04835554,0.0001007395,0.00004342508,0.00004755055,9.416418e-7,0.001045064],"genre_scores_gemma":[0.9987767,0.00001711292,0.000004930675,0.0001555425,0.0006451437,0.000001567783,0.00002020349,2.221805e-7,0.0003786136],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08691701,"threshold_uncertainty_score":0.9350833,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03150941875166875,"score_gpt":0.2813356173383293,"score_spread":0.2498261985866606,"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."}}