{"id":"W2946284567","doi":"10.1145/3314994.3325089","title":"OATMEAL","year":2019,"lang":"en","type":"article","venue":"","topic":"Teaching and Learning Programming","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Victoria","funders":"","keywords":"Debugging; Computer science; Software engineering; Software testing; Code (set theory); Curriculum; Programming language; Competition (biology); Multimedia; Software; Pedagogy; Psychology","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0001198855,0.00002709483,0.00003088321,0.00001777283,0.00002560406,0.00007180133,0.0002546531,0.00001229597,0.00004905347],"category_scores_gemma":[0.00000629219,0.00002122565,0.00001698237,0.00006498488,0.000002770953,0.0001125497,0.00005810682,0.00006937557,0.001445959],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000004692715,"about_ca_system_score_gemma":0.000008058995,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002448643,"about_ca_topic_score_gemma":5.975502e-7,"domain_scores_codex":[0.9996791,0.00001707725,0.00003343699,0.0001091005,0.00006575169,0.00009545931],"domain_scores_gemma":[0.999737,0.00002049605,0.000009552753,0.0002043706,0.000005784693,0.00002279963],"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":[2.841613e-7,0.00001069177,0.008824995,0.000001923245,0.000002335765,0.000001390636,0.0004787553,0.00005391288,0.00006896072,0.2059912,0.0004697999,0.7840958],"study_design_scores_gemma":[0.0001579,0.00007291838,0.009616799,0.000004648238,5.336217e-7,0.00001036201,0.00003117003,0.01386223,0.0001256336,0.0003282036,0.9756792,0.000110358],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.08128388,0.00002295584,0.8656796,0.0009847981,0.0004056856,0.00004574872,1.258997e-8,0.0007027426,0.05087456],"genre_scores_gemma":[0.8368936,1.347114e-7,0.1536682,0.0001679245,0.0000133942,5.96649e-7,1.047745e-7,0.000001547782,0.009254462],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9752094,"threshold_uncertainty_score":0.9993315,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.005241045796087448,"score_gpt":0.2101322493687713,"score_spread":0.2048912035726838,"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."}}