{"id":"W4366453408","doi":"10.3138/cjpe.020.002","title":"Ensuring Quality for Evaluation: Lessons from Auditors","year":2005,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Audit; Documentation; Quality assurance; Quality (philosophy); Variety (cybernetics); Quality audit; Business; Point (geometry); Accounting; Process management; Computer science; Marketing","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04357931,0.0001481295,0.0003109899,0.0005469818,0.0003680295,0.000511034,0.0005448051,0.0001111621,0.005333714],"category_scores_gemma":[0.009494307,0.0001220437,0.0002324066,0.0005267738,0.00007481853,0.0009457702,0.00001157692,0.0001895385,0.0001188194],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009757869,"about_ca_system_score_gemma":0.007917928,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000676141,"about_ca_topic_score_gemma":0.05250241,"domain_scores_codex":[0.9930303,0.001025715,0.001484704,0.0003007757,0.003828621,0.0003298784],"domain_scores_gemma":[0.9898871,0.000920682,0.001197414,0.0003998704,0.007075246,0.0005196914],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002184201,0.000036794,0.006979414,0.000001957089,0.00003687148,3.428305e-7,0.0009794647,0.01248903,0.00005091636,0.0004190751,0.008313416,0.9706709],"study_design_scores_gemma":[0.003894454,0.0003493298,0.2503506,0.00006065141,0.000289267,0.000009107907,0.001872875,0.3154173,0.0004467474,0.03356598,0.3934413,0.000302405],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9556951,0.001343331,0.006545775,0.02529365,0.004085224,0.00314509,0.00008641076,0.00002065648,0.003784802],"genre_scores_gemma":[0.9851113,0.000006715522,0.0122423,0.0002713952,0.001932087,0.0002183765,0.00005230441,0.00001248831,0.0001530397],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9703684,"threshold_uncertainty_score":0.9988492,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6231916254204312,"score_gpt":0.6243646498296325,"score_spread":0.001173024409201306,"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."}}