{"id":"W4380271342","doi":"10.1177/15553434231171484","title":"Does It MultiMatch? What Scanpath Comparison Tells us About Task Performance in Teams","year":2023,"lang":"en","type":"article","venue":"Journal of Cognitive Engineering and Decision Making","topic":"Human-Automation Interaction and Safety","field":"Psychology","cited_by":16,"is_retracted":false,"has_abstract":true,"ca_institutions":"Carleton University","funders":"","keywords":"Teamwork; Workload; Task (project management); Computer science; Context (archaeology); Similarity (geometry); Artificial intelligence; Machine learning; Engineering","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":[0.0006743203,0.0001392075,0.0003023571,0.0006644368,0.00006896613,0.0001338517,0.00009449826,0.00008030864,0.0003323765],"category_scores_gemma":[0.0003227297,0.00009757499,0.00007496269,0.0003057203,0.00002468379,0.0005006446,0.00003382421,0.0004024254,0.0001347357],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004668402,"about_ca_system_score_gemma":0.00001608063,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000001645678,"about_ca_topic_score_gemma":0.00000449101,"domain_scores_codex":[0.9986452,0.00004173043,0.0006775686,0.0001516245,0.000272333,0.0002114919],"domain_scores_gemma":[0.9982139,0.001193252,0.0002638156,0.00007856515,0.0001769504,0.00007346839],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0006144146,0.000145391,0.03990433,0.00005540111,0.0001127351,0.0001981784,0.01740441,0.04411967,0.0002556582,0.00008903164,0.001314561,0.8957862],"study_design_scores_gemma":[0.002011481,0.0001736382,0.7203553,0.005702092,0.00002570081,0.0001322867,0.01249795,0.25165,0.0000911899,0.0001116034,0.006987548,0.0002611714],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9713817,0.0003044017,0.02511619,0.00006237882,0.002347076,0.00008361258,0.00000354745,0.00004445074,0.0006566842],"genre_scores_gemma":[0.9982653,0.0003068931,0.0009701085,0.0001228626,0.0001642885,0.000004832601,0.000001698098,0.00001811791,0.000145955],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.895525,"threshold_uncertainty_score":0.3978993,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02030495061285501,"score_gpt":0.3666193155963371,"score_spread":0.3463143649834821,"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."}}