{"id":"W2037599414","doi":"10.1111/capa.12103","title":"What metrics? On the utility of measuring the performance of policy research: An illustrative case and alternative from Employment and Social Development Canada","year":2015,"lang":"en","type":"article","venue":"Canadian Public Administration","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa; Institute on Governance","funders":"","keywords":"Government (linguistics); Task (project management); Object (grammar); Performance measurement; Focus (optics); Computer science; Research Object; State (computer science); Policy development; Public policy; Management science; Operations research; Public economics; Econometrics; Economics; Sociology; Engineering; Political science; Public administration; Artificial intelligence; Regional science; Economic growth; Management","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006095131,0.00009210308,0.0001347941,0.0002859987,0.000582212,0.0003481236,0.0002890359,0.0000411112,0.00008626018],"category_scores_gemma":[0.001481514,0.0000550513,0.00001194121,0.0006598191,0.0002909972,0.0006959672,0.0000443587,0.0001612866,0.000001056662],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004242572,"about_ca_system_score_gemma":0.01725063,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.4118102,"about_ca_topic_score_gemma":0.9860116,"domain_scores_codex":[0.9969639,0.000583665,0.0004228909,0.00023603,0.001581875,0.0002116815],"domain_scores_gemma":[0.9972369,0.0008061992,0.000221302,0.000255843,0.001177121,0.0003026114],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0003221947,0.0002604212,0.1258408,0.00003709673,0.0002664644,0.00008845038,0.1467291,0.00004018459,0.00007609311,0.125809,0.01255965,0.5879706],"study_design_scores_gemma":[0.001307413,0.001385552,0.3434657,0.00007108044,0.00002550351,0.0000847474,0.499196,0.05450509,0.01234996,0.005361256,0.08174032,0.0005074193],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9699546,0.00003824412,0.00001328267,0.028773,0.0001102608,0.000277283,0.0000568331,0.000001721822,0.0007748306],"genre_scores_gemma":[0.999461,0.00001039046,0.00004112415,0.0002949133,0.00006512504,0.00002052233,0.00001451881,0.000003548763,0.00008883586],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5874631,"threshold_uncertainty_score":0.9883206,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6470003141615459,"score_gpt":0.4995877186147932,"score_spread":0.1474125955467527,"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."}}