{"id":"W3208537877","doi":"10.1097/jom.0000000000002293","title":"Evidence for Validity and Reliability, and Development of Performance Standards and Cut-Scores for Job-Related Tests of Physical Aptitude for Structural Firefighters","year":2021,"lang":"en","type":"article","venue":"Journal of Occupational and Environmental Medicine","topic":"Occupational Health and Performance","field":"Health Professions","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Northern Alberta Institute of Technology","funders":"","keywords":"Aptitude; Reliability (semiconductor); Psychology; Applied psychology; Test validity; Psychometrics; Clinical psychology; Developmental psychology","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":[],"consensus_categories":[],"category_scores_codex":[0.0007969458,0.0001043875,0.0003315034,0.00004835343,0.0002904407,0.000002005579,0.00003562634,0.00005918429,0.00001928178],"category_scores_gemma":[0.000401034,0.00007291894,0.00002616701,0.00003081791,0.0002313197,0.0001857762,0.00004014649,0.0001155739,4.265904e-8],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007488434,"about_ca_system_score_gemma":0.0002766303,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000003837716,"about_ca_topic_score_gemma":0.00001039555,"domain_scores_codex":[0.9987149,0.00003968343,0.0006548464,0.000137212,0.0003123178,0.0001410764],"domain_scores_gemma":[0.9977062,0.001520944,0.0004229021,0.00005029014,0.0001747485,0.0001249407],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002974637,0.00004356969,0.9773272,0.002828429,0.00003166114,2.691293e-7,0.002650097,0.00003114924,0.002593617,0.0001006948,0.000187646,0.01123097],"study_design_scores_gemma":[0.002846702,0.001261098,0.9897683,0.0009913748,0.00005356698,0.00001446809,0.0003684043,0.002602209,0.0005754267,0.0004703891,0.0009767606,0.00007132592],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9967923,0.001234734,0.0006508756,0.0004221252,0.0001330297,0.0004554932,0.0003063406,0.000001364074,0.000003699046],"genre_scores_gemma":[0.9916896,0.000966935,0.00700232,0.00008703746,0.0001537529,0.00002655606,0.00004495764,0.000006116162,0.00002277902],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01244102,"threshold_uncertainty_score":0.2973548,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1306249037519791,"score_gpt":0.454148332761945,"score_spread":0.3235234290099659,"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."}}