{"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":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.0007133965,0.00006094857,0.0000820348,0.00005677515,0.001519219,0.0001031719,0.0003796505,0.00005809854,0.000477057],"category_scores_gemma":[0.007852879,0.00003981066,0.00002729135,0.0002378627,0.0003206329,0.0001768688,0.000308796,0.0001223967,0.00004942257],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001956786,"about_ca_system_score_gemma":0.00001337968,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001154015,"about_ca_topic_score_gemma":0.00008850075,"domain_scores_codex":[0.9995742,0.000005047252,0.0001277094,0.0001074474,0.00008742319,0.00009815083],"domain_scores_gemma":[0.9989336,0.0003630276,0.0001691747,0.0003953367,0.0001375786,0.000001283625],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.000002352396,0.00008717331,0.07258383,0.00001276118,0.00001340464,2.060142e-7,0.00004355941,3.500046e-7,0.00006634833,0.9088272,0.01755303,0.0008097189],"study_design_scores_gemma":[0.0001172069,0.00001742278,0.6750976,0.00009936237,0.00001827506,2.748279e-7,0.001256832,0.0002694179,0.001624201,0.2913866,0.029889,0.0002237669],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.2474372,0.000001471076,0.0002440815,0.004832203,0.0001846211,0.0001564985,7.159865e-7,0.00006305413,0.7470801],"genre_scores_gemma":[0.9934244,7.217708e-7,0.0002349139,0.0007100159,0.00008464715,0.00002305971,3.601727e-7,0.000004427497,0.005517497],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7459871,"threshold_uncertainty_score":0.9997807,"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."}}