{"id":"W2582325117","doi":"10.1145/3013499.3013500","title":"The Compound Nature of Novice Programming Assessments","year":2017,"lang":"en","type":"article","venue":"","topic":"Teaching and Learning Programming","field":"Computer Science","cited_by":65,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Multitude; Computer science; Mathematics education; Data science; Psychology; Epistemology","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":["sts","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.0007718751,0.00007590388,0.00008714201,0.00001970669,0.001350532,0.001117064,0.002021324,0.00006179226,0.000001161941],"category_scores_gemma":[0.0001424254,0.00004701446,0.00004820216,0.00005097038,0.00009172534,0.0003297724,0.0003960412,0.0004287717,0.000006991476],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001148202,"about_ca_system_score_gemma":0.0000353688,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001526463,"about_ca_topic_score_gemma":0.00003835036,"domain_scores_codex":[0.999159,0.00005931394,0.0001289199,0.0001892016,0.000246176,0.0002174157],"domain_scores_gemma":[0.998523,0.0001490277,0.0002147281,0.00100981,0.00006328633,0.00004012715],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000001626959,0.0000279419,0.005809128,0.000006439393,0.00001542557,0.00000284116,0.0001941232,0.000004488733,0.0000306793,0.06628109,0.0001195315,0.9275067],"study_design_scores_gemma":[0.0002850588,0.0001084666,0.03867726,0.00004200054,0.000006415714,0.00001459458,0.0001466869,0.006301234,0.0002650626,0.0004721041,0.9535171,0.0001639761],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.08558065,0.0005004116,0.876883,0.007668265,0.001940912,0.0004179973,4.229636e-7,0.0006326998,0.02637565],"genre_scores_gemma":[0.8347077,0.00000215191,0.1642883,0.00004602118,0.00003325519,0.000004318563,2.99572e-7,0.000004044367,0.0009139277],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9533976,"threshold_uncertainty_score":0.9999496,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02101629712874874,"score_gpt":0.3464425089190803,"score_spread":0.3254262117903316,"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."}}