{"id":"W7032976869","doi":"","title":"Oopsy-daisy:Failure stories in quantitative evaluation studies for visualizations","year":2014,"lang":"en","type":"article","venue":"","topic":"Animal Disease Management and Epidemiology","field":"Agricultural and Biological Sciences","cited_by":1,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Visualization; Interpretation (philosophy); Feature (linguistics); Quantitative analysis (chemistry); Data visualization","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05752194,0.001808644,0.00122361,0.0041599,0.001661704,0.00674428,0.002974894,0.002687984,0.0190236],"category_scores_gemma":[0.2835253,0.001070961,0.001413204,0.00364032,0.002446791,0.01227386,0.006072262,0.003629334,0.001663181],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001929526,"about_ca_system_score_gemma":0.001890988,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002545592,"about_ca_topic_score_gemma":0.004507911,"domain_scores_codex":[0.9472171,0.04199684,0.003104255,0.001540121,0.005674866,0.000466769],"domain_scores_gemma":[0.4278424,0.5303687,0.01000005,0.02090123,0.009590107,0.001297482],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.002201457,0.0007020785,0.02799121,0.009664127,0.0007942939,0.00120988,0.04042194,0.04829352,0.007641608,0.2605638,0.1623259,0.4381903],"study_design_scores_gemma":[0.0005797259,0.0007319531,0.009505701,0.002950628,0.0002627684,0.0008136718,0.009378993,0.481765,0.01132681,0.3207477,0.1615712,0.0003658774],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03215659,0.000804896,0.9149299,0.005264045,0.0003996275,0.0008774313,0.01099626,0.0228808,0.01169045],"genre_scores_gemma":[0.3658023,0.0004320175,0.6166389,0.0005585438,0.0001950515,0.002376904,0.005464457,0.005614121,0.002917839],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9424781,"threshold_uncertainty_score":0.3042088,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2164326620908712,"score_gpt":0.4263012254625288,"score_spread":0.2098685633716576,"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."}}