{"id":"W1505229351","doi":"10.1111/j.1360-0443.2011.03375.x","title":"Commentary on Brennan <i>et al</i>. (2011): Towards more interpretable evaluations","year":2011,"lang":"en","type":"letter","venue":"Addiction","topic":"Substance Abuse Treatment and Outcomes","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Centre for Addiction and Mental Health","funders":"","keywords":"Psychological intervention; Comparability; SAFER; Intervention (counseling); Psychology; Aggression; Outcome (game theory); Behavior change; Injury prevention; Applied psychology; Poison control; Medicine; Clinical psychology; Social psychology; Medical emergency; Psychiatry; Computer security; Computer science","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":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0001696446,0.0003879626,0.0004724349,0.0002720825,0.0001054781,0.0000215285,0.0001207448,0.0003338677,0.004571022],"category_scores_gemma":[0.00002056756,0.000315742,0.0002639763,0.0001048739,0.00006493051,0.0001651908,0.00003084955,0.001066691,0.0009734649],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003515719,"about_ca_system_score_gemma":0.0001000315,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002542893,"about_ca_topic_score_gemma":0.00004516209,"domain_scores_codex":[0.9982599,0.00009893252,0.0003211187,0.0004567594,0.0005300545,0.0003332038],"domain_scores_gemma":[0.998899,0.0000912366,0.0001649385,0.0006636514,0.0001079105,0.00007325855],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001270701,0.000122623,0.001646131,0.00001424649,0.0005141178,0.0001244098,0.0006405495,7.160979e-7,0.00001668022,0.000004957433,0.9869117,0.009876871],"study_design_scores_gemma":[0.001685274,0.0004756259,0.004814453,0.0004681858,0.001066572,0.0000398522,0.0000606014,0.00004122673,0.0003113075,0.00005722953,0.9907197,0.0002599801],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.002431658,0.0007941567,0.0001017881,0.9797762,0.001704079,0.0007971358,0.00008390322,0.0002965188,0.01401458],"genre_scores_gemma":[0.006573713,0.0006514748,0.0002279041,0.9767396,0.001085185,0.0001912415,0.006364422,0.00008766747,0.008078826],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.00961689,"threshold_uncertainty_score":0.9999295,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03420351465345901,"score_gpt":0.3141281979719266,"score_spread":0.2799246833184676,"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."}}