{"id":"W2078446277","doi":"10.1016/j.knosys.2010.07.008","title":"Advanced empirical testing","year":2010,"lang":"en","type":"article","venue":"Knowledge-Based Systems","topic":"Software System Performance and Reliability","field":"Computer Science","cited_by":12,"is_retracted":false,"has_abstract":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Computer 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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001119163,0.0002483397,0.0003510073,0.0001545338,0.0002797595,0.0002154805,0.001078054,0.0002026544,0.000009046234],"category_scores_gemma":[0.0005480626,0.0001955271,0.0001141572,0.0009297472,0.00008777854,0.0003806442,0.0001390253,0.0004338255,0.0008002855],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007258642,"about_ca_system_score_gemma":0.0004641922,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003695092,"about_ca_topic_score_gemma":0.00003087978,"domain_scores_codex":[0.9978746,0.0001565104,0.0005288647,0.0006278658,0.0003253958,0.0004867926],"domain_scores_gemma":[0.9971361,0.0006618868,0.0001633248,0.001331234,0.0004759832,0.0002314731],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00004945334,0.00137185,0.7085546,0.002358913,0.00007548903,0.00007482637,0.002925224,0.003085812,0.05515214,0.007126648,0.01340857,0.2058165],"study_design_scores_gemma":[0.002746817,0.0004923322,0.0428686,0.0006122668,0.0000216418,0.0001642809,0.0001139363,0.7480156,0.01131443,0.0003298214,0.1918543,0.001465986],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7016909,0.0007668773,0.2555438,0.0003718301,0.0165778,0.001021795,0.000005345265,0.002185733,0.02183589],"genre_scores_gemma":[0.9864899,3.723828e-7,0.01239784,0.00006401558,0.0005028628,0.0001201335,0.000002311432,0.00001982692,0.0004027467],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7449298,"threshold_uncertainty_score":0.9999777,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03136029153661333,"score_gpt":0.2997289248272138,"score_spread":0.2683686332906005,"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."}}