{"id":"W4410303778","doi":"10.1007/s43441-025-00787-x","title":"Benchmarking Medical Information Services Beyond the Unsolicited Requests: A phactMI Benchmarking Survey","year":2025,"lang":"en","type":"article","venue":"Therapeutic Innovation & Regulatory Science","topic":"Pharmaceutical industry and healthcare","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"EVERSANA (Canada)","funders":"","keywords":"Benchmarking; Medical information; Medical education; Health care; Information needs; Medicine; Business; Knowledge management; Family medicine; Marketing; Computer science; Political science; World Wide Web","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":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01342828,0.0002881953,0.0002689587,0.000693533,0.001952511,0.0001938709,0.001276515,0.0005794258,0.00104024],"category_scores_gemma":[0.0002500758,0.0002254976,0.00004835731,0.006682746,0.001593281,0.001366563,0.0002381294,0.001851924,0.00005478037],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003167239,"about_ca_system_score_gemma":0.00148418,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001032696,"about_ca_topic_score_gemma":0.00004964057,"domain_scores_codex":[0.9958821,0.0008926157,0.0009974033,0.0004219209,0.001007529,0.0007984235],"domain_scores_gemma":[0.9969071,0.0008074024,0.0004034727,0.0004826685,0.001179622,0.0002198063],"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":[0.0004685748,0.0002144147,0.1081286,0.0002576188,0.0001960473,0.000006447387,0.00286877,0.0004343592,0.01456398,0.2625332,0.002620901,0.6077071],"study_design_scores_gemma":[0.002083843,0.0001519663,0.645053,0.0002452893,0.000121664,0.00004018567,0.0007249199,0.1051264,0.01870297,0.01084512,0.2162,0.0007046488],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9717801,0.0004257438,0.001054913,0.009844736,0.003065524,0.000757846,0.00003487831,0.0001914557,0.01284482],"genre_scores_gemma":[0.933064,0.00007502505,0.00005054515,0.06619003,0.0002949014,0.00004663835,0.0001012939,0.000009091432,0.0001684553],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6070025,"threshold_uncertainty_score":0.9998729,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2154787304488383,"score_gpt":0.5208728671645962,"score_spread":0.3053941367157579,"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."}}