{"id":"W1587172963","doi":"10.1111/j.1365-2753.2006.00730.x","title":"External validity: the neglected dimension in evidence ranking","year":2006,"lang":"en","type":"article","venue":"Journal of Evaluation in Clinical Practice","topic":"Heart Failure Treatment and Management","field":"Medicine","cited_by":56,"is_retracted":false,"has_abstract":true,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"External validity; Internal validity; Evidence-based medicine; Ranking (information retrieval); Context (archaeology); Evidence-based practice; Medicine; Psychology; Randomized controlled trial; Social psychology; Alternative medicine; Computer science; Surgery","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":[0.01770172,0.00008397499,0.0002832713,0.0001612667,0.00003909513,0.00003453328,0.00007002657,0.00007620658,0.0001526207],"category_scores_gemma":[0.02926955,0.00005219895,0.0001237779,0.0003438081,0.00003703111,0.0005852035,0.00002974562,0.0006408754,0.00002501029],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000217543,"about_ca_system_score_gemma":0.0002607097,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001427829,"about_ca_topic_score_gemma":0.0001005861,"domain_scores_codex":[0.995118,0.001994689,0.001462271,0.0001400591,0.001146347,0.0001386219],"domain_scores_gemma":[0.9901914,0.00814147,0.0007905857,0.000183797,0.0006437789,0.00004897043],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01122982,0.006724327,0.7603486,0.00009547294,0.0002616129,0.002298952,0.0007426239,0.005459219,0.003040369,0.001007769,0.04551083,0.1632804],"study_design_scores_gemma":[0.00722514,0.0009203228,0.9616671,0.001099905,0.0007397771,0.000292062,0.0001730931,0.006399713,0.00005894255,0.002188654,0.01915988,0.0000754791],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9090447,0.001341886,0.001004319,0.08603944,0.0006348058,0.0006497095,7.95841e-8,0.000006165184,0.001278866],"genre_scores_gemma":[0.9896265,0.0004658047,0.00773717,0.001455568,0.0005802401,0.000007643663,8.399539e-7,0.00000705053,0.0001191518],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2013185,"threshold_uncertainty_score":0.9789073,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2081572965841139,"score_gpt":0.5058346538587319,"score_spread":0.297677357274618,"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."}}