{"id":"W4408703448","doi":"10.1109/tse.2025.3553383","title":"Do Experts Agree About Smelly Infrastructure?","year":2025,"lang":"en","type":"article","venue":"IEEE Transactions on Software Engineering","topic":"Tracheal and airway disorders","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo; McGill University","funders":"","keywords":"Computer science; Software engineering; Data science; Engineering management; Engineering","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":[0.00003848824,0.0002096889,0.0002361641,0.0002687369,0.00009142328,0.00002314615,0.00008373864,0.0001323861,0.0002125839],"category_scores_gemma":[0.00002142713,0.0001968894,0.0001552668,0.0003627556,0.00002534427,0.00007453611,9.248537e-7,0.0003313351,0.00002301569],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008696467,"about_ca_system_score_gemma":0.00006519981,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001329787,"about_ca_topic_score_gemma":0.000004264796,"domain_scores_codex":[0.9990872,0.000006333228,0.0002119348,0.000257159,0.0001694172,0.0002680256],"domain_scores_gemma":[0.9994708,0.00007443906,0.00001784544,0.0002771481,0.00004386457,0.0001159577],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0003051039,0.0005161546,0.002696524,0.0009083511,0.0005542276,0.00007008236,0.001132794,0.3972576,0.008089727,0.0003289959,0.004296629,0.5838438],"study_design_scores_gemma":[0.02488359,0.002342373,0.3354044,0.008752333,0.002040919,0.0004806549,0.00231612,0.04858698,0.2373462,0.0007087343,0.3320333,0.005104459],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1017291,0.0004643663,0.8954013,0.000166618,0.0009983052,0.000214913,0.00001370137,0.0005806663,0.0004310588],"genre_scores_gemma":[0.9894193,0.00009598998,0.00869091,0.0003337018,0.00005430127,0.00004971454,0.000003562748,0.00003311692,0.001319352],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8876902,"threshold_uncertainty_score":0.8028916,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.006346365505415781,"score_gpt":0.2373352788788933,"score_spread":0.2309889133734775,"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."}}