{"id":"W7117297114","doi":"10.48550/arxiv.2512.20713","title":"How to Do STEM Outreach Evaluation -- Recommendations Based on a Review of Self Evaluation Tools in Canadian STEM Outreach Programs","year":2025,"lang":"","type":"preprint","venue":"arXiv (Cornell University)","topic":"Career Development and Diversity","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Memorial University of Newfoundland","funders":"","keywords":"Outreach; Grassroots; Toolbox; Program evaluation; Accreditation; Equity (law)","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.01094544,0.0006008013,0.0009308588,0.001662125,0.0006555718,0.0003436374,0.001362355,0.0005819015,0.0004822344],"category_scores_gemma":[0.0003553616,0.0007725546,0.0004573511,0.004389689,0.0001294938,0.0006748302,0.0004356497,0.0006195197,0.00005613593],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.01165408,"about_ca_system_score_gemma":0.01163311,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.1207115,"about_ca_topic_score_gemma":0.47054,"domain_scores_codex":[0.9929196,0.002589749,0.0007102955,0.001758886,0.001137158,0.000884346],"domain_scores_gemma":[0.9942228,0.0003331463,0.0007476988,0.001092794,0.002868331,0.0007352934],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0003012607,0.00186746,0.3631016,0.006045934,0.0007124784,0.00005786995,0.02809643,0.03219875,0.000001520395,0.01829888,0.007027594,0.5422902],"study_design_scores_gemma":[0.01570047,0.001419898,0.1207328,0.1125139,0.01234891,0.000001603369,0.1100889,0.4452983,0.00008510317,0.003400071,0.1706724,0.007737565],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6264228,0.003482674,0.003414793,0.04503758,0.007672952,0.06396632,0.002011763,0.0005077476,0.2474834],"genre_scores_gemma":[0.9922803,0.002378623,0.0002223932,0.0004593463,0.00005728471,0.00005874392,0.0007664735,0.00001700972,0.003759839],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5345526,"threshold_uncertainty_score":0.9994726,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2363010666598031,"score_gpt":0.2823270138478078,"score_spread":0.04602594718800473,"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."}}