{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2301938,0.003209711,0.005310522,0.03115697,0.007780312,0.01424285,0.008944014,0.004685579,0.003085214],"category_scores_gemma":[0.2844551,0.002211325,0.004349614,0.0318921,0.006813505,0.0139628,0.007818152,0.005944929,0.001339911],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.09394842,"about_ca_system_score_gemma":0.3936684,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.7787161,"about_ca_topic_score_gemma":0.8964629,"domain_scores_codex":[0.876269,0.05703898,0.02429496,0.003737395,0.03227957,0.006380137],"domain_scores_gemma":[0.5606528,0.1153112,0.0221798,0.006728628,0.2783122,0.01681527],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0002601291,0.0002523805,0.01065404,0.1133212,0.0003817821,0.0003033461,0.01392624,0.0006799368,0.0007082297,0.004609134,0.1157521,0.7391514],"study_design_scores_gemma":[0.000336577,0.0002412663,0.02512484,0.5734826,0.001935052,0.0002102985,0.02196592,0.0009605128,0.001150464,0.002938496,0.3712929,0.0003611912],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.008292628,0.7402478,0.01253307,0.2028532,0.007086636,0.01060907,0.002476641,0.0008325996,0.01506817],"genre_scores_gemma":[0.05883836,0.7556177,0.1253644,0.03702248,0.0008765279,0.01399132,0.002810179,0.0003541385,0.00512492],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9060516,"threshold_uncertainty_score":0.9493081,"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."}}