{"id":"W2165295550","doi":"10.1093/reseval/rvs041","title":"Accountability, performance assessment, and evaluation: Policy pressures and responses from research councils","year":2013,"lang":"en","type":"article","venue":"Research Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":30,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Accountability; Public administration; Political science; Library science; Higher education; Sociology; Management; Law; Computer science; 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","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.2249294,0.0002326165,0.0003545237,0.001638426,0.00155351,0.002507815,0.000801266,0.0002015294,0.01101138],"category_scores_gemma":[0.02493386,0.0001783446,0.0000466495,0.002504644,0.0009609738,0.003053465,0.0005682752,0.0008954044,0.0008199532],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00123662,"about_ca_system_score_gemma":0.006424922,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002660922,"about_ca_topic_score_gemma":0.0005345074,"domain_scores_codex":[0.9490788,0.01641608,0.001134457,0.001420878,0.03094723,0.001002531],"domain_scores_gemma":[0.9596035,0.0136839,0.0002330193,0.001437485,0.02464267,0.0003993895],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0001664923,0.0001631363,0.2386146,0.00003251234,0.00004731108,4.07505e-7,0.00378333,0.0004038191,0.003610151,0.001199819,0.01997693,0.7320015],"study_design_scores_gemma":[0.000879486,0.0002377975,0.6272569,0.00003965411,0.00001645337,0.000002440876,0.002357947,0.3080627,0.0002092489,0.0535362,0.007266427,0.0001347649],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9679695,0.00126615,0.00006588069,0.009967044,0.0001508346,0.00363794,0.00003206094,0.00003198267,0.01687862],"genre_scores_gemma":[0.9945518,0.0009756837,0.0008102734,0.0001391138,0.0004280552,0.001558044,0.00004922464,0.0000209204,0.001466831],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7318667,"threshold_uncertainty_score":0.999958,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8015169397706693,"score_gpt":0.7149106275809038,"score_spread":0.08660631218976544,"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."}}