{"id":"W2796847126","doi":"10.2307/1602249","title":"Test Construction and Validation: Case Description of Constructing and Validating a Test of English for Teaching Purposes","year":2002,"lang":"en","type":"article","venue":"Canadian Journal of Education / Revue canadienne de l éducation","topic":"Educational Technology and Assessment","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"","keywords":"Test (biology); Computer science; Mathematics education; Psychology; Geology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0870705,0.001910233,0.00109494,0.003517585,0.007610024,0.004342943,0.006315097,0.009494188,0.001303871],"category_scores_gemma":[0.3019675,0.001815822,0.002298258,0.002117356,0.008243567,0.004542251,0.007481335,0.005588123,0.0009405364],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005460678,"about_ca_system_score_gemma":0.01082471,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01230123,"about_ca_topic_score_gemma":0.01456405,"domain_scores_codex":[0.8613049,0.09486338,0.007925243,0.006119606,0.02420039,0.005586445],"domain_scores_gemma":[0.6339769,0.2713165,0.01130255,0.04528944,0.03397075,0.004143886],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001253804,0.004093885,0.1259615,0.002093436,0.0002473078,0.03776207,0.3945599,0.0135781,0.05437537,0.02327866,0.006168311,0.3366277],"study_design_scores_gemma":[0.000457687,0.008414944,0.1109308,0.005183905,0.001082544,0.06897266,0.2159552,0.1227527,0.3183564,0.03430593,0.1123075,0.001279894],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.7509844,0.0003601908,0.2321873,0.002725575,0.000155294,0.003048501,0.000188724,0.0008404471,0.009509568],"genre_scores_gemma":[0.7674003,0.0001495346,0.2260491,0.0006719942,0.00003201032,0.001249951,0.0003082537,0.0004893086,0.003649545],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.0870705,"threshold_uncertainty_score":0.4604783,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04615243925074911,"score_gpt":0.2699627301624387,"score_spread":0.2238102909116896,"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."}}