{"id":"W1214597323","doi":"10.26481/dis.20150617dc","title":"Teacher-based reactivity to provincial large-scale assessment in Canada","year":2015,"lang":"en","type":"dissertation","venue":"","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Scale (ratio); Accountability; Test (biology); Mathematics education; Psychology; Medical education; Pedagogy; Political science; Geography; Medicine; Cartography","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003065944,0.0003026141,0.0004929103,0.0004152937,0.00008813821,0.0002073124,0.0007714856,0.0001354858,0.001634103],"category_scores_gemma":[0.0008278362,0.0002378406,0.00008072755,0.0007381487,0.00000613838,0.0002279567,0.00005958239,0.0004628381,0.00008546362],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.003878195,"about_ca_system_score_gemma":0.0436542,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.8812168,"about_ca_topic_score_gemma":0.9993703,"domain_scores_codex":[0.9939597,0.0002049896,0.0008709602,0.0008181054,0.003709591,0.0004366376],"domain_scores_gemma":[0.997551,0.0004301729,0.0003739938,0.000589638,0.0007383156,0.0003168754],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.0002060379,0.001634637,0.2587627,0.00003800318,0.00002504841,0.00001199226,0.00146345,0.000218938,0.001217649,0.0004848528,0.6639003,0.07203645],"study_design_scores_gemma":[0.0008030302,0.0001846608,0.8143956,0.00005669706,0.00001974988,2.119321e-7,0.04178636,0.002414812,0.001052798,0.003484603,0.1351131,0.0006882972],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8990607,0.00001003698,0.0002178382,0.002851708,0.002151949,0.0008944778,0.00009397716,0.00001659111,0.09470271],"genre_scores_gemma":[0.9294133,4.261636e-7,0.001870038,0.0009863421,0.0002444338,0.0002851786,0.0007509585,0.00002514696,0.06642424],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5556329,"threshold_uncertainty_score":0.9999458,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06305222848576099,"score_gpt":0.428410135310017,"score_spread":0.3653579068242561,"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."}}