{"id":"W1825023912","doi":"","title":"Using Large-scale Assessments to Determine Student Grades","year":2013,"lang":"en","type":"article","venue":"Canadian Journal of Education / Revue canadienne de l éducation","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Prince Edward Island","funders":"","keywords":"Scale (ratio); Tweaking; Psychology; Mathematics education; Medical education; Guideline; Medicine; Computer science; Geography; Cartography","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001162252,0.0001290111,0.000181874,0.0006748595,0.0005627159,0.0002861729,0.0004267264,0.00009137313,0.001353152],"category_scores_gemma":[0.0003776776,0.0001505039,0.00007345294,0.0005533885,0.00005887051,0.0006050591,0.00000767841,0.0001770004,0.00004173917],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.005516578,"about_ca_system_score_gemma":0.02314793,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.2062877,"about_ca_topic_score_gemma":0.8681188,"domain_scores_codex":[0.9982913,0.0002251677,0.0004592147,0.0001819126,0.000153338,0.0006890318],"domain_scores_gemma":[0.9959635,0.00009778446,0.0003481032,0.0001938103,0.001233416,0.002163378],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000004918692,0.000378112,0.457523,0.00005149161,0.00007521546,0.00000721798,0.4524091,0.0002081403,0.0007742292,0.02058264,0.05107509,0.01691092],"study_design_scores_gemma":[0.0001812481,0.0001048925,0.3727518,0.0001737159,0.00007132582,0.00007131381,0.3357944,0.00007720701,0.0000304175,0.004922883,0.2854899,0.0003307958],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9773775,0.0001176767,0.0004408384,0.01117903,0.005781356,0.0003549171,0.000009211303,0.000005492499,0.004733915],"genre_scores_gemma":[0.9864582,0.00001302258,0.005620984,0.001312547,0.001801105,0.00004343091,0.00002014714,0.00001911512,0.00471149],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.661831,"threshold_uncertainty_score":0.9995598,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1356944071166592,"score_gpt":0.4296797860607543,"score_spread":0.2939853789440952,"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."}}