{"id":"W6949826937","doi":"10.5281/zenodo.16794826","title":"ResearchBox 490, 'When Metrics Matter: Impact of Elicitation Metric', https://researchbox.org/490","year":2025,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Decision-Making and Behavioral Economics","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Backup; Expert elicitation; Metric (unit); Measure (data warehouse); Preference elicitation; Selection (genetic algorithm)","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01034247,0.001185283,0.001600968,0.003093323,0.001621433,0.005952274,0.001966021,0.002674411,0.7385787],"category_scores_gemma":[0.1134587,0.0009782517,0.0008167747,0.003973791,0.0007827958,0.007181096,0.003725744,0.001752203,0.5675197],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00222705,"about_ca_system_score_gemma":0.003841303,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003355972,"about_ca_topic_score_gemma":0.005767012,"domain_scores_codex":[0.9949157,0.001526617,0.0005950628,0.0005811502,0.0021023,0.0002792351],"domain_scores_gemma":[0.921898,0.04594259,0.003713403,0.01054802,0.01492067,0.002977355],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001881642,0.0000445757,0.0002780717,0.0007052496,0.000006976423,0.00001738542,0.0001054662,0.00004857353,0.0002079349,0.001399918,0.9694018,0.02759586],"study_design_scores_gemma":[0.0004906264,0.00007241179,0.002951468,0.001258859,0.00003015459,0.00006359489,0.0002642154,0.0005097756,0.00217118,0.01104293,0.9810837,0.00006108801],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"other","genre_scores_codex":[0.003453002,0.001659002,0.02332455,0.01636277,0.00433369,0.00205596,0.4950696,0.1162195,0.337522],"genre_scores_gemma":[0.05479877,0.003056215,0.08666929,0.01077499,0.002609854,0.009597329,0.3161285,0.1794085,0.3369567],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.9896575,"threshold_uncertainty_score":0.3728858,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1557771206082407,"score_gpt":0.4069713172112631,"score_spread":0.2511941966030223,"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."}}