{"id":"W4417447051","doi":"10.1080/0969594x.2025.2602452","title":"The validity of the assessment for learning measurement instrument for Ethiopian middle school mathematics teachers","year":2025,"lang":"en","type":"article","venue":"Assessment in Education Principles Policy and Practice","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Measure (data warehouse); Reliability (semiconductor); Test validity; Test (biology)","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":[],"consensus_categories":[],"category_scores_codex":[0.02509119,0.0004753883,0.0004735617,0.002713248,0.001206333,0.001863692,0.0008074236,0.0004404683,0.00087939],"category_scores_gemma":[0.06866992,0.0002678621,0.000421516,0.002594801,0.001236281,0.001388591,0.001523689,0.0007811438,0.0002731461],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001988469,"about_ca_system_score_gemma":0.004006575,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003937587,"about_ca_topic_score_gemma":0.004312937,"domain_scores_codex":[0.9867355,0.007481258,0.001346752,0.0007289325,0.003150366,0.0005573183],"domain_scores_gemma":[0.9450843,0.03502186,0.006934823,0.002879687,0.009147325,0.0009320038],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002035547,0.0003441249,0.7720174,0.0003713926,0.0001165423,0.0001514893,0.009709666,0.001080524,0.00165944,0.003847423,0.0009497346,0.2095487],"study_design_scores_gemma":[0.00008673143,0.00119927,0.9408668,0.001172407,0.0001170436,0.0005812123,0.02479991,0.00885389,0.003771646,0.004806842,0.01360343,0.0001407946],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9835914,0.0005334151,0.007943384,0.0004125445,0.0000697621,0.0003015671,0.000329566,0.00003538586,0.006782939],"genre_scores_gemma":[0.9832169,0.0002697231,0.01549633,0.000083182,0.00001635191,0.0003455656,0.0001946539,0.000008955735,0.0003682141],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02509119,"threshold_uncertainty_score":0.1326964,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.596064082416965,"score_gpt":0.5664535415641934,"score_spread":0.02961054085277159,"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."}}