{"id":"W1557717928","doi":"10.1002/ev.20082","title":"The Value in Validity","year":2014,"lang":"en","type":"article","venue":"New Directions for Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"123 Certification (Canada)","funders":"","keywords":"Beauty; Prioritization; Economic Justice; Value (mathematics); Sociology; Balance (ability); Management science; Psychology; Social psychology; Epistemology; Computer science; Political science; Law; Economics","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02828205,0.00008381571,0.0001184255,0.0002003183,0.0005597453,0.0002651981,0.0002929553,0.0000515813,0.0004453527],"category_scores_gemma":[0.01400155,0.00005413893,0.00009171481,0.0007903934,0.00003311063,0.0003258729,0.00002427962,0.00008651624,0.0002515431],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001737388,"about_ca_system_score_gemma":0.0003940238,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009400039,"about_ca_topic_score_gemma":0.001564777,"domain_scores_codex":[0.9963982,0.0008966677,0.0005428375,0.0002901654,0.001685257,0.0001868754],"domain_scores_gemma":[0.994442,0.004180217,0.0001902834,0.0004646098,0.0006610256,0.00006189734],"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":[0.00003208042,0.0000354521,0.006722787,0.000001101709,0.000005441278,9.353223e-9,0.0004275869,0.01422823,0.00009164034,0.01937688,0.03017574,0.928903],"study_design_scores_gemma":[0.0007418859,0.00006068935,0.05916897,0.000003837933,0.00001399135,5.561759e-7,0.0001307301,0.4134066,0.0002153973,0.1348695,0.3913291,0.00005873568],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5854608,0.0003399578,0.2047027,0.03869582,0.009478688,0.004196056,0.00001243521,0.0001903395,0.1569231],"genre_scores_gemma":[0.9894294,0.00003332132,0.001398457,0.0002727378,0.0003654762,0.0003094481,0.00001088535,0.000008599464,0.008171696],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9288443,"threshold_uncertainty_score":0.9943039,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3622601772662026,"score_gpt":0.5450118219863205,"score_spread":0.1827516447201178,"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."}}