{"id":"W3181957689","doi":"10.1145/3460319.3464837","title":"Empirical evaluation of smart contract testing: what is the best choice?","year":2021,"lang":"en","type":"article","venue":"","topic":"Blockchain Technology Applications and Security","field":"Computer Science","cited_by":83,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Computer science; Vulnerability (computing); Empirical research; Computer security; Risk analysis (engineering); Data science","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3366262,0.00228694,0.00417279,0.00515047,0.00153763,0.009226035,0.005653507,0.006283563,0.005701358],"category_scores_gemma":[0.7137919,0.00101494,0.002294988,0.007179043,0.01017717,0.01576656,0.003240795,0.005110884,0.001827681],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002951587,"about_ca_system_score_gemma":0.004137733,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001917079,"about_ca_topic_score_gemma":0.00195127,"domain_scores_codex":[0.5952786,0.3522699,0.01886417,0.008122182,0.02311135,0.002353743],"domain_scores_gemma":[0.09457504,0.8078464,0.03656972,0.02854007,0.0283893,0.004079376],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01029879,0.00491284,0.3644423,0.01623313,0.01103787,0.0006351533,0.003643504,0.02701361,0.001768872,0.06680976,0.05558559,0.4376185],"study_design_scores_gemma":[0.006107068,0.01527414,0.2458559,0.04933567,0.008565531,0.002396756,0.01657202,0.1869434,0.008150702,0.3548789,0.1047532,0.001166857],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4094616,0.09691511,0.3559217,0.09183875,0.003107838,0.004844591,0.008839638,0.001091308,0.02797949],"genre_scores_gemma":[0.9219673,0.007412932,0.05574305,0.005610025,0.001105509,0.002365785,0.004656997,0.0003218608,0.0008164839],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6633738,"threshold_uncertainty_score":0.818058,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07999998757489113,"score_gpt":0.3409563399381531,"score_spread":0.260956352363262,"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."}}