{"id":"W1534140708","doi":"10.1002/ev.20088","title":"House With a View: Validity and Evaluative Argument","year":2014,"lang":"en","type":"article","venue":"New Directions for Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"123 Certification (Canada)","funders":"","keywords":"CLARITY; Argument (complex analysis); Style (visual arts); Epistemology; Narrative; Evaluation methods; Psychology; Computer science; Sociology; Linguistics; Philosophy; History","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":[0.1004092,0.0006014818,0.001210546,0.002569561,0.006011734,0.01902502,0.002526195,0.007074974,0.006318588],"category_scores_gemma":[0.2555495,0.0007013882,0.0007399375,0.002142657,0.04504205,0.02666344,0.006939893,0.01051741,0.0008116884],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008304209,"about_ca_system_score_gemma":0.007916982,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003517228,"about_ca_topic_score_gemma":0.002354083,"domain_scores_codex":[0.8421661,0.1307555,0.002637188,0.00410498,0.01819811,0.002138176],"domain_scores_gemma":[0.7050767,0.2628883,0.00541187,0.01021667,0.01483076,0.001575751],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00002512148,0.00002312422,0.0001770521,0.0001096693,0.00001696981,0.00007809785,0.006656816,0.0001303073,0.00005472607,0.967012,0.01663014,0.009085909],"study_design_scores_gemma":[0.00004446672,0.00005653198,0.000658752,0.001617282,0.00003361566,0.00008609361,0.01256525,0.001368654,0.000664743,0.8410231,0.1418375,0.00004384234],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"methods","genre_scores_codex":[0.01973745,0.01664108,0.06048966,0.5924214,0.007104952,0.0003522162,0.00009211645,0.0001178921,0.3030433],"genre_scores_gemma":[0.8483149,0.008094476,0.03570234,0.07275525,0.004866804,0.0008826451,0.00009471743,0.0003112113,0.02897766],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8995908,"threshold_uncertainty_score":0.5310211,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.240140513715607,"score_gpt":0.5046303264375429,"score_spread":0.2644898127219358,"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."}}