{"id":"W4400335907","doi":"10.1145/3649405.3659507","title":"A Framework for Evaluating the Impact of Production Quality on Coding Demonstration Videos","year":2024,"lang":"en","type":"article","venue":"","topic":"Industrial Vision Systems and Defect Detection","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"The Scarborough Hospital; University of Toronto","funders":"","keywords":"Computer science; Production (economics); Coding (social sciences); Quality (philosophy); Mathematics; Statistics","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":[],"consensus_categories":[],"category_scores_codex":[0.001277717,0.00007380908,0.00009924736,0.00005861899,0.00007019224,0.00005627042,0.00002965801,0.00008737622,0.00002312751],"category_scores_gemma":[0.0005317232,0.00004363746,0.0001224373,0.0002115873,0.00000788112,0.00008653758,0.000002997983,0.0001402545,0.000007707452],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000903082,"about_ca_system_score_gemma":0.00002508042,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006847631,"about_ca_topic_score_gemma":0.000003495576,"domain_scores_codex":[0.9993451,0.00005629544,0.0002595927,0.0001109525,0.0001385873,0.00008944577],"domain_scores_gemma":[0.9993554,0.0004171808,0.00003220077,0.0001289358,0.00005196586,0.00001435345],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002156662,0.00002712786,0.0001973388,0.0004758812,0.0002529102,2.562229e-7,0.001310594,0.3582089,0.3220682,0.05448406,0.002893815,0.2598653],"study_design_scores_gemma":[0.0002020513,0.0009520468,0.001819691,0.0006710234,0.00003829549,0.000008255799,0.0004742185,0.8416728,0.1457564,0.00803134,0.0001718188,0.0002020795],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6930007,0.00005786528,0.3037419,0.00005277139,0.001294502,0.0005614268,0.000005170756,0.0002215073,0.001064163],"genre_scores_gemma":[0.9986835,0.000002424761,0.0007169196,0.000002400469,0.0004623497,0.00005063137,0.000001728027,0.00001298419,0.00006707083],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4834639,"threshold_uncertainty_score":0.1779484,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1603194633588637,"score_gpt":0.4434181242685702,"score_spread":0.2830986609097065,"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."}}