{"id":"W2990790246","doi":"10.1186/s12916-019-1436-0","title":"Accuracy in detecting inadequate research reporting by early career peer reviewers using an online CONSORT-based peer-review tool (COBPeer) versus the usual peer-review process: a cross-sectional diagnostic study","year":2019,"lang":"en","type":"article","venue":"BMC Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":78,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ottawa Public Health; Ottawa Hospital","funders":"Cancer Research UK","keywords":"Medicine; Peer review; Consolidated Standards of Reporting Trials; Randomized controlled trial; Family medicine; Cross-sectional study; MEDLINE; Internal medicine; Pathology","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","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5733609,0.0007486138,0.006124126,0.0006220257,0.0005758468,0.0008532447,0.003226677,0.0001650007,0.005371324],"category_scores_gemma":[0.9039847,0.0003524544,0.001074228,0.00582518,0.0003613331,0.0007221039,0.0003159517,0.00153182,0.0005346829],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002996707,"about_ca_system_score_gemma":0.001173136,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001091167,"about_ca_topic_score_gemma":0.0008741412,"domain_scores_codex":[0.8792062,0.03203441,0.0322356,0.003578991,0.05151568,0.001429085],"domain_scores_gemma":[0.5292979,0.238897,0.06745928,0.02664214,0.1358523,0.001851256],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00008772742,0.0006968355,0.9164265,0.008842527,0.0002274123,0.00008254209,0.001214034,0.0005962452,0.00004436097,0.00001190227,0.05279605,0.01897387],"study_design_scores_gemma":[0.01108786,0.003966934,0.5784677,0.0658624,0.004104889,0.0002899143,0.0206916,0.03394708,0.00002475389,0.0002794167,0.2787904,0.002486991],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9026169,0.07761295,0.0004766273,0.008726929,0.0009147456,0.009172891,0.00004586794,0.00002318346,0.0004099573],"genre_scores_gemma":[0.9850627,0.001649438,0.0005944466,0.001988657,0.0005854981,0.0006616514,0.0002131962,0.00007866239,0.009165738],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3499083,"threshold_uncertainty_score":0.9998928,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8886671091862027,"score_gpt":0.6380346204566281,"score_spread":0.2506324887295747,"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."}}