{"id":"W2740136032","doi":"10.1145/3077136.3080644","title":"On the Reusability of \"Living Labs\" Test Collections","year":2017,"lang":"en","type":"article","venue":"","topic":"Innovative Approaches in Technology and Social Development","field":"Business, Management and Accounting","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Reusability; Computer science; Test (biology); Software engineering; Operating system; Software; Biology; Ecology","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08953076,0.001460307,0.001085357,0.007374173,0.002392505,0.005259897,0.002932416,0.002103228,0.001469408],"category_scores_gemma":[0.4326583,0.0008014724,0.001359918,0.004913806,0.003684363,0.006453421,0.004341844,0.002544161,0.0008464238],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002207379,"about_ca_system_score_gemma":0.001578813,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005677895,"about_ca_topic_score_gemma":0.008573216,"domain_scores_codex":[0.8678505,0.07159956,0.01120996,0.01230922,0.03457653,0.002454312],"domain_scores_gemma":[0.2956926,0.5097388,0.03633307,0.09711698,0.05777166,0.00334692],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.004195927,0.004857465,0.2703379,0.003440884,0.002803331,0.001394441,0.0321335,0.02649104,0.04823962,0.004970075,0.01778725,0.5833485],"study_design_scores_gemma":[0.0007556988,0.0115559,0.6751794,0.001521225,0.002087756,0.003466657,0.01362218,0.110158,0.1171148,0.009971296,0.05341682,0.001150186],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9287835,0.001253194,0.05649746,0.0003680361,0.0001782549,0.001668943,0.00228554,0.002384746,0.006580259],"genre_scores_gemma":[0.9574742,0.0001901169,0.03234363,0.0002258982,0.00008297888,0.001174178,0.006293141,0.0008657572,0.001350029],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08953076,"threshold_uncertainty_score":0.4734896,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0362506119348016,"score_gpt":0.2466168726120512,"score_spread":0.2103662606772496,"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."}}