{"id":"W4387246794","doi":"10.1145/3626570.3626589","title":"Teaching Software Performance Evaluation to Undergrads: Lessons Learned and Challenges","year":2023,"lang":"en","type":"article","venue":"ACM SIGMETRICS Performance Evaluation Review","topic":"Software Engineering Techniques and Practices","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Engineering management; Software; Computer science; Engineering; Operating system","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":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0172031,0.0002625149,0.0003385419,0.000791278,0.0003987004,0.0001621084,0.001107076,0.00009753284,0.00003146248],"category_scores_gemma":[0.01282048,0.0002475653,0.00006257062,0.002962083,0.00001855004,0.001936555,0.0006727296,0.000358771,0.0002272291],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001819224,"about_ca_system_score_gemma":0.0001859956,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000007989805,"about_ca_topic_score_gemma":0.00000173188,"domain_scores_codex":[0.9963908,0.000428076,0.0004734804,0.0006648127,0.001607547,0.0004352485],"domain_scores_gemma":[0.996668,0.001100885,0.0002492025,0.001302024,0.0005341842,0.0001456891],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000001675734,0.0000144244,0.0003266596,0.0009414108,0.00001391816,4.34499e-7,0.0001957503,0.00210143,0.000005275997,0.0006705683,0.001321269,0.9944072],"study_design_scores_gemma":[0.0005297153,0.000519507,0.05092981,0.00343751,0.000227569,0.0000286653,0.00003423603,0.787227,0.0001028921,0.002026473,0.1541114,0.0008252112],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.1290472,0.5432792,0.1942443,0.1174601,0.001855961,0.007200085,0.000009045531,0.005257894,0.001646282],"genre_scores_gemma":[0.2584931,0.6773537,0.06217106,0.0009648084,0.0001059566,0.0007410154,0.00003282718,0.00003631551,0.0001012232],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.993582,"threshold_uncertainty_score":0.9999977,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.264157883154723,"score_gpt":0.4154940041769574,"score_spread":0.1513361210222344,"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."}}