{"id":"W2007266934","doi":"10.1177/154193120504900323","title":"Decision-Centered Testing (DCT): Evaluating Joint Human-Computer Cognitive Work","year":2005,"lang":"en","type":"article","venue":"Proceedings of the Human Factors and Ergonomics Society Annual Meeting","topic":"Human-Automation Interaction and Safety","field":"Psychology","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Defence Research and Development Canada","funders":"","keywords":"Usability; Computer science; Cognition; Discrete cosine transform; Decision support system; Test (biology); Machine learning; Artificial intelligence; Human–computer interaction; 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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0009791014,0.0003076368,0.0003869585,0.0000694581,0.001152889,0.0001611283,0.0003005966,0.0001729361,0.0002332347],"category_scores_gemma":[0.0002735809,0.0002463185,0.0003330184,0.0002036868,0.0001930168,0.0003712478,0.0003321373,0.0004755275,0.00001846615],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000131169,"about_ca_system_score_gemma":0.00001781613,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005463263,"about_ca_topic_score_gemma":0.000005735288,"domain_scores_codex":[0.9979442,0.00003168374,0.0008744966,0.0004936859,0.0002504848,0.0004054817],"domain_scores_gemma":[0.9979616,0.0004248609,0.0008539953,0.0001193735,0.0005297521,0.00011037],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004145777,0.0009608071,0.544246,0.0003110608,0.001218627,5.876109e-7,0.3037046,0.0004223003,0.02306159,0.007680033,0.03449113,0.08348875],"study_design_scores_gemma":[0.002845879,0.0003612889,0.9319901,0.001763012,0.0001911613,0.00001727276,0.05285314,0.003577549,0.003245983,0.0007954916,0.001469816,0.0008892599],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9936373,0.00005761395,0.00007303368,0.0000967624,0.0004951474,0.0003182392,0.00002222524,0.000111502,0.005188171],"genre_scores_gemma":[0.9939331,0.000003646107,0.004607333,0.0003046397,0.0005832774,0.00001853629,0.000005683673,0.00004195973,0.0005018],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3877442,"threshold_uncertainty_score":0.9999989,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07211093901151318,"score_gpt":0.3469451116592162,"score_spread":0.274834172647703,"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."}}