{"id":"W3163618964","doi":"10.17605/osf.io/chm2k","title":"ARDiff: An Equivalence Checking Framework","year":2020,"lang":"en","type":"article","venue":"OSF Preprints (OSF Preprints)","topic":"Intensive Care Unit Cognitive Disorders","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Equivalence (formal languages); Computer science; Mathematics; Calculus (dental); Discrete mathematics","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","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.001359693,0.0004121335,0.0006218157,0.0001375016,0.0001750168,0.0000869889,0.0007526423,0.0003334278,0.3365312],"category_scores_gemma":[0.04828257,0.0004588814,0.0003267875,0.0004331368,0.0003154631,0.0002996627,0.0009675099,0.001193307,0.5943669],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001835237,"about_ca_system_score_gemma":0.000229404,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004977286,"about_ca_topic_score_gemma":0.000003416247,"domain_scores_codex":[0.9954656,0.0003966826,0.0005553094,0.002317765,0.0006287721,0.0006358786],"domain_scores_gemma":[0.993522,0.0004668489,0.0002137009,0.002596302,0.002458976,0.0007421876],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.005801674,0.002692218,0.3652151,0.002330207,0.002365973,0.002721984,0.04899247,0.001011811,0.2659352,0.01878164,0.1178382,0.1663136],"study_design_scores_gemma":[0.01039204,0.0001190705,0.3122286,0.004195788,0.002198265,0.001234682,0.03425852,0.01201347,0.2802242,0.02754073,0.310344,0.005250657],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.2852795,0.000005196938,0.02613288,0.007708573,0.000368867,0.001438339,0.00002423176,0.0004824775,0.67856],"genre_scores_gemma":[0.9533528,0.00006878777,0.004443788,0.02856593,0.0003075991,0.0001085023,0.0000577573,0.0001000188,0.01299481],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6680733,"threshold_uncertainty_score":0.9997863,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03276542947209719,"score_gpt":0.3029208496694737,"score_spread":0.2701554201973765,"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."}}