{"id":"W2147607381","doi":"10.1177/1098214007309280","title":"The Evaluation of Large Research Initiatives","year":2008,"lang":"en","type":"article","venue":"American Journal of Evaluation","topic":"scientometrics and bibliometrics research","field":"Decision Sciences","cited_by":144,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Program evaluation; Agency (philosophy); Government (linguistics); Work (physics); Political science; Management science; Public relations; Public administration; Sociology; Engineering; Social science","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","bibliometrics"],"consensus_categories":["metaresearch","bibliometrics"],"category_scores_codex":[0.3656344,0.00007352234,0.0002771633,0.02057371,0.0006159107,0.0002704363,0.00123374,0.00003090312,0.0003782494],"category_scores_gemma":[0.2249136,0.00004000981,0.000151052,0.09118454,0.0009350633,0.0007370567,0.0001744744,0.0003724133,0.00004785818],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002743431,"about_ca_system_score_gemma":0.002218674,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002523443,"about_ca_topic_score_gemma":0.00001130628,"domain_scores_codex":[0.9306285,0.009501756,0.001593362,0.0002975611,0.05746344,0.0005153461],"domain_scores_gemma":[0.9140736,0.01479045,0.001600276,0.0005007997,0.06885068,0.0001842276],"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":[0.0001298367,0.0002203134,0.05034644,8.969214e-7,0.00005282047,0.00000363811,0.002031767,0.000862361,0.0009045146,0.0008260244,0.008424918,0.9361964],"study_design_scores_gemma":[0.001321333,0.001622692,0.8919392,0.00001646788,0.00003838969,0.00005981162,0.01014377,0.06581713,0.001415102,0.02350648,0.004031598,0.00008802626],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9891057,0.001834449,0.001439018,0.001238122,0.0003090158,0.0002812435,0.000004714444,0.00000148896,0.005786228],"genre_scores_gemma":[0.9986893,0.0007605782,0.000323713,0.00002475659,0.0001222435,0.000008076996,8.201211e-7,0.000005414658,0.00006509508],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9361085,"threshold_uncertainty_score":0.9905273,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8816684527440397,"score_gpt":0.727468180018336,"score_spread":0.1542002727257037,"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."}}