{"id":"W4412658976","doi":"10.1080/14992027.2025.2537693","title":"Validation of the clinical assistant for research and learning (CARL) for pure-tone audiometric procedures","year":2025,"lang":"en","type":"article","venue":"International Journal of Audiology","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Process Simulations Limited (Canada); Western University","funders":"","keywords":"Audiology; Tone (literature); Pure tone; Audiometry; Psychology; Medicine; Speech recognition; Computer science; Hearing loss; Linguistics; Philosophy","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"],"consensus_categories":[],"category_scores_codex":[0.002582625,0.00004557958,0.0002329335,0.0004267081,0.00006773195,0.00001211379,0.0001337569,0.00008492143,0.00001558201],"category_scores_gemma":[0.01296012,0.00002876084,0.00014459,0.0002620475,0.0001541148,0.00003576251,0.00004238148,0.0003022044,2.252005e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000492084,"about_ca_system_score_gemma":0.0002500188,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000001075948,"about_ca_topic_score_gemma":0.000001303331,"domain_scores_codex":[0.9988619,0.0001396504,0.0005468019,0.00009648598,0.000255451,0.00009967222],"domain_scores_gemma":[0.9943957,0.003394848,0.0003132766,0.00004734551,0.001811381,0.00003748913],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.006719705,0.0003735285,0.8869457,0.0001451593,0.001207152,0.000009287308,0.000227,0.0006543421,0.004806051,0.02000524,0.001749729,0.07715716],"study_design_scores_gemma":[0.01134327,0.002061273,0.9339913,0.0004445369,0.0001339514,0.0001372866,0.0003526568,0.001243892,0.004582352,0.01760918,0.02803137,0.00006891896],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.98462,0.0002593915,0.006602234,0.006851413,0.0007008595,0.0002506904,0.000004346987,0.000003579341,0.0007074737],"genre_scores_gemma":[0.9978932,0.00007838257,0.001011537,0.0001898332,0.0003007747,0.00000758049,0.000004855574,0.000004386728,0.0005094775],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07708824,"threshold_uncertainty_score":0.9953541,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1687758687450233,"score_gpt":0.5292294074968484,"score_spread":0.3604535387518252,"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."}}