{"id":"W1987538235","doi":"10.1145/1297846.1297882","title":"Open framework for conformance testing via scenarios","year":2007,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Carleton University","funders":"","keywords":"Conformance testing; Computer science; Software engineering; Software testing; Formal specification; Conformance checking; Software; Reliability engineering; Programming language; Operating system; Engineering; Compatibility (geochemistry); Business process; Standardization","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":[],"consensus_categories":[],"category_scores_codex":[0.00113811,0.0001026484,0.0001226669,0.00006611507,0.000213471,0.0003243107,0.001767533,0.00007084855,0.000004142383],"category_scores_gemma":[0.001510528,0.0000890097,0.00002712504,0.0003918041,0.00002725486,0.0004438362,0.0006022264,0.0001102076,0.00002023721],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002523077,"about_ca_system_score_gemma":0.00005554132,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001471401,"about_ca_topic_score_gemma":0.000006006562,"domain_scores_codex":[0.9990405,0.000006801831,0.0002134235,0.0002763172,0.0001241649,0.0003388039],"domain_scores_gemma":[0.9973463,0.001798658,0.00008997955,0.0005274285,0.0001549834,0.00008264105],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000009501151,0.00004747703,0.008841584,0.00002241218,0.000006921788,0.000008117354,0.0002524531,0.00001558335,0.000102268,0.2539323,0.01492445,0.7218369],"study_design_scores_gemma":[0.0002245538,0.0002338981,0.002989776,0.0001521985,0.000002962472,0.00004777287,0.000004588319,0.1435035,0.004723148,0.8421811,0.005595436,0.0003409891],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0004090953,0.00001725854,0.9823155,0.0002745795,0.0001764979,0.0003544235,3.480036e-7,0.007991269,0.008460968],"genre_scores_gemma":[0.1902809,2.409062e-7,0.8078678,0.001606727,0.00005632923,0.00002243196,5.251823e-7,0.000007749342,0.0001572996],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7214959,"threshold_uncertainty_score":0.362971,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07472432542418293,"score_gpt":0.3517893958620555,"score_spread":0.2770650704378726,"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."}}