{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003782361,0.0004345594,0.000576362,0.002960213,0.0006627366,0.001993066,0.0005687697,0.0008242206,0.002890977],"category_scores_gemma":[0.03774302,0.0002647283,0.0005112355,0.002217362,0.001377914,0.004660274,0.001265612,0.0006539807,0.0004451507],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005258329,"about_ca_system_score_gemma":0.0003713389,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002136195,"about_ca_topic_score_gemma":0.002624308,"domain_scores_codex":[0.9977455,0.0008828223,0.0001452328,0.0006539546,0.0004335273,0.0001389621],"domain_scores_gemma":[0.9767209,0.0135617,0.005024219,0.002175818,0.001567669,0.0009495587],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002311343,0.0002595039,0.6447547,0.000456057,0.0005814856,0.0002837012,0.01043663,0.003796521,0.008615995,0.004758163,0.002441077,0.3213048],"study_design_scores_gemma":[0.00004173501,0.001049019,0.9480668,0.00017757,0.0001503163,0.0006143179,0.006869708,0.01205075,0.002438452,0.02458364,0.003829959,0.0001276939],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9721562,0.001047853,0.01998523,0.0007340987,0.00008609162,0.0000307937,0.0003077387,0.0001613927,0.005490474],"genre_scores_gemma":[0.9940975,0.0002564249,0.004805342,0.0001171965,0.00005333568,0.00003127517,0.0002037146,0.00004708043,0.0003882523],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003782361,"threshold_uncertainty_score":0.02000326,"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."}}