{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03872744,0.0001402789,0.0002441197,0.0002868176,0.0007114109,0.0003301754,0.0006645796,0.00008103788,0.000006017118],"category_scores_gemma":[0.3479875,0.00008251938,0.0001037405,0.001147127,0.0001068319,0.0003339389,0.0002158804,0.0003802642,4.486896e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004825642,"about_ca_system_score_gemma":0.003323106,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001932812,"about_ca_topic_score_gemma":0.00007542888,"domain_scores_codex":[0.9957099,0.001794594,0.0009756769,0.0003464481,0.0008991339,0.0002742465],"domain_scores_gemma":[0.892214,0.1045765,0.001400005,0.0007401768,0.0009897684,0.00007948784],"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":[0.00004936677,0.0007329295,0.09627533,0.0001606188,0.0001032397,3.339102e-8,0.002108771,0.0005693522,0.0003483622,0.665193,0.001789519,0.2326695],"study_design_scores_gemma":[0.0006510887,0.0001417461,0.2626672,0.0001786879,0.0000782771,0.000003444517,0.04758152,0.003221957,0.0002317051,0.1607387,0.5243675,0.0001382016],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.288953,0.000792162,0.3400139,0.2666278,0.004841681,0.006748653,0.00002496612,0.0000589729,0.09193895],"genre_scores_gemma":[0.6983185,0.0003085405,0.2967338,0.001180318,0.0001380889,0.0006762997,0.00000231742,0.00001061968,0.002631494],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.522578,"threshold_uncertainty_score":0.9898324,"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."}}