{"id":"W4406212572","doi":"10.1145/3711838","title":"Comparative Structured Observation","year":2025,"lang":"en","type":"article","venue":"ACM Transactions on Computer-Human Interaction","topic":"Usability and User Interface Design","field":"Computer Science","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada; Agence Nationale de la Recherche","keywords":"Computer science","routes":{"ca_aff":true,"ca_fund":true,"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":[0.05550256,0.0009122043,0.0006554211,0.004379034,0.004912193,0.003233422,0.002546672,0.001968913,0.01293895],"category_scores_gemma":[0.09520646,0.0006230058,0.0007849603,0.00513979,0.004862013,0.004275757,0.006566134,0.0023476,0.001763942],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004352958,"about_ca_system_score_gemma":0.006571972,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002337647,"about_ca_topic_score_gemma":0.004690238,"domain_scores_codex":[0.9349975,0.05108954,0.002289335,0.005010231,0.005369202,0.001244227],"domain_scores_gemma":[0.880752,0.06733523,0.004835995,0.02334215,0.0224322,0.001302388],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0008606914,0.0008240395,0.009374257,0.004262014,0.0001119889,0.000993023,0.4586349,0.002383739,0.02291191,0.1099328,0.03711903,0.3525916],"study_design_scores_gemma":[0.000786913,0.002478872,0.01804386,0.004058152,0.000178695,0.001297383,0.2693908,0.01076939,0.02756021,0.09859966,0.5665683,0.0002678454],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.192305,0.001724437,0.6240472,0.004812988,0.001016505,0.02309456,0.003555355,0.001143818,0.1483002],"genre_scores_gemma":[0.4545147,0.00106925,0.4941241,0.001361039,0.0002141727,0.02991802,0.001282913,0.00043181,0.01708393],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.05550256,"threshold_uncertainty_score":0.2935291,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06864231019348375,"score_gpt":0.3400075220299694,"score_spread":0.2713652118364856,"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."}}