{"id":"W2005408543","doi":"10.1191/0265532204lt292oa","title":"Test decisions over time: tracking validity","year":2004,"lang":"en","type":"article","venue":"Language Testing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":69,"is_retracted":false,"has_abstract":true,"ca_institutions":"Carleton University","funders":"","keywords":"Test (biology); Psychology; Sample (material); Active listening; Test validity; Applied psychology; Social psychology; Mathematics education; Psychometrics; Developmental psychology","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0004003863,0.0001371782,0.000142537,0.00006679029,0.0006815977,0.0002795259,0.0001496499,0.00003576185,0.001591122],"category_scores_gemma":[0.00457736,0.0001198853,0.00006053765,0.00006109421,0.00006603329,0.0001996025,0.00004755471,0.0003634228,0.0005117817],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004464213,"about_ca_system_score_gemma":0.0000301434,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001115818,"about_ca_topic_score_gemma":0.0002256885,"domain_scores_codex":[0.9990923,0.00004395917,0.0001912657,0.000207482,0.0001821669,0.000282794],"domain_scores_gemma":[0.9982728,0.001321398,0.00008957538,0.0001997668,0.00005006103,0.00006644046],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001355728,0.0006795053,0.03277876,0.0000868864,0.00006419853,0.000793916,0.6610945,0.002074301,0.03906832,0.02491748,0.003123549,0.235305],"study_design_scores_gemma":[0.01943282,0.00356865,0.2340919,0.01532846,0.001250235,0.001230249,0.2057081,0.03342386,0.0125521,0.04692489,0.4131491,0.01333958],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8992832,0.000153095,0.000195114,0.00009807546,0.000161516,0.00006820215,0.00001140488,0.0005819982,0.09944735],"genre_scores_gemma":[0.9915507,5.549055e-7,0.002472898,0.0002869051,0.0008213656,0.000003526356,0.00001312138,0.00003343383,0.004817429],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4553863,"threshold_uncertainty_score":0.9993216,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1004204092065655,"score_gpt":0.2939022965666889,"score_spread":0.1934818873601234,"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."}}