{"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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01470984,0.0001630329,0.0002359726,0.0002837742,0.0004739558,0.0002613241,0.0001637112,0.00006029147,0.000967174],"category_scores_gemma":[0.003387637,0.0001144784,0.00006992483,0.0006268728,0.00006107021,0.0005457156,0.00003137769,0.00008689269,0.0001660828],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000166844,"about_ca_system_score_gemma":0.0004715519,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005176118,"about_ca_topic_score_gemma":0.0005579006,"domain_scores_codex":[0.9956893,0.0007325164,0.0004859223,0.0005260939,0.002364073,0.0002020958],"domain_scores_gemma":[0.9962776,0.001326908,0.0003010812,0.0004531097,0.001491778,0.0001494988],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001423616,0.0001200493,0.008983283,0.000007454547,0.00003833132,4.086496e-8,0.001670273,0.003960565,0.0002208531,0.002459656,0.01871305,0.9636841],"study_design_scores_gemma":[0.004800286,0.001216396,0.1073725,0.00003609455,0.0003053381,0.000008932052,0.0006307045,0.4839411,0.001335116,0.06402384,0.3359998,0.0003299913],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6927229,0.0002786646,0.2771509,0.006789342,0.001234768,0.00379451,0.0000172652,0.0002592919,0.01775236],"genre_scores_gemma":[0.9857093,0.00005422368,0.009424304,0.0004004726,0.0002791543,0.0005754763,0.00002101126,0.00002483789,0.003511203],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9633541,"threshold_uncertainty_score":0.9999461,"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."}}