{"id":"W4407384694","doi":"10.3102/01623737241311537","title":"Measuring Grading Standards at High Schools: A Methodological Contribution, an Example, and Some Policy Implications","year":2025,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Higher Education Learning Practices","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Wilfrid Laurier University","funders":"Canadian Institutes of Health Research; University of Alberta","keywords":"Grading (engineering); Academic standards; Mathematics education; Policy analysis; Higher education; Econometrics; Psychology; Political science; Economics; Economic growth; Public administration; Engineering","routes":{"ca_aff":true,"ca_fund":true,"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":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.009614435,0.0001470931,0.0002774922,0.001118429,0.002131317,0.0003595375,0.000181708,0.0001298525,0.0007471397],"category_scores_gemma":[0.01825696,0.0001524565,0.00008135653,0.003160891,0.000270999,0.0007671216,0.00005666643,0.0001836117,0.000010272],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00144054,"about_ca_system_score_gemma":0.005138886,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03066875,"about_ca_topic_score_gemma":0.003612088,"domain_scores_codex":[0.9949667,0.003081406,0.0003901523,0.0004906548,0.000770348,0.0003007132],"domain_scores_gemma":[0.995616,0.001942897,0.0002857046,0.0002911877,0.001564486,0.0002996946],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00001649671,0.00008736865,0.08591696,0.00000659839,0.0003608128,2.21452e-8,0.007365124,0.0004106984,0.000103292,0.8992026,0.001135398,0.005394621],"study_design_scores_gemma":[0.0003155639,0.00001507169,0.7717223,0.00001124736,0.0008693979,0.000001100308,0.002033581,0.0003895997,0.00001784639,0.1487566,0.07569472,0.0001729887],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6317106,0.0009176367,0.002753463,0.3606406,0.000173709,0.0004460443,0.00005506771,0.00007038638,0.003232522],"genre_scores_gemma":[0.9913523,0.0004064797,0.002550978,0.001753022,0.0008895954,0.0001999677,0.000219902,0.000006980098,0.002620752],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.750446,"threshold_uncertainty_score":0.9991678,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2785140261361238,"score_gpt":0.5528497145082416,"score_spread":0.2743356883721178,"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."}}