{"id":"W7155759947","doi":"","title":"SNAP vs. FSA: Competency-Based Numeracy Assessments Count","year":2021,"lang":"en","type":"dissertation","venue":"KU ScholarWorks (The University of Kansas)","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Numeracy; Context (archaeology); Christian ministry; Academic achievement; Sample (material); Cohort; Standardized test; Achievement test","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.000841963,0.0002520231,0.0004015486,0.0001658297,0.00173557,0.0001602202,0.001084279,0.0004529236,0.005840254],"category_scores_gemma":[0.00008008419,0.0002705744,0.000297735,0.0005743335,0.0002485651,0.0004639341,0.00004157146,0.0009320199,0.0001130272],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003549954,"about_ca_system_score_gemma":0.00594891,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004647189,"about_ca_topic_score_gemma":0.02522468,"domain_scores_codex":[0.9973125,0.0005770733,0.0002534923,0.0004236893,0.001073876,0.0003593748],"domain_scores_gemma":[0.9976962,0.0004442841,0.0004621419,0.0004048061,0.0008342908,0.0001583281],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001784495,0.004348613,0.1974768,0.00138525,0.002763957,0.0002312633,0.184155,0.001363374,0.0006088766,0.4364974,0.1249731,0.04441196],"study_design_scores_gemma":[0.0009722925,0.00007980317,0.0822546,0.0006607996,0.0005381334,6.717875e-7,0.2110779,0.0001410765,0.00005931922,0.001827641,0.7016712,0.0007164949],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.4802598,0.001902812,0.00106474,0.01457223,0.006415305,0.001380539,0.0001562652,0.0001862619,0.4940621],"genre_scores_gemma":[0.8664915,0.0009769036,0.003935822,0.0005695201,0.0005373787,0.000003550582,0.002968808,0.00004515213,0.1244714],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5766982,"threshold_uncertainty_score":0.9999747,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02177446102862877,"score_gpt":0.3400671295468901,"score_spread":0.3182926685182613,"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."}}