{"id":"W2083551948","doi":"10.1186/1472-6920-12-20","title":"Expert validation of fit-for-purpose guidelines for designing programmes of assessment","year":2012,"lang":"en","type":"article","venue":"BMC Medical Education","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":60,"is_retracted":false,"has_abstract":true,"ca_institutions":"The Wilson Centre; University of Toronto","funders":"","keywords":"Rubric; Competence (human resources); Brainstorming; Computer science; Context (archaeology); Judgement; Management science; Set (abstract data type); Process management; Medical education; Psychology; Medicine; Engineering; Artificial intelligence; Pedagogy","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2735832,0.001747804,0.001635111,0.009762095,0.00255746,0.004761588,0.004103712,0.003568309,0.002476292],"category_scores_gemma":[0.4908795,0.001103968,0.001925128,0.005737986,0.003191891,0.005056248,0.004859352,0.002710249,0.001308346],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00677823,"about_ca_system_score_gemma":0.0168336,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002608847,"about_ca_topic_score_gemma":0.005816915,"domain_scores_codex":[0.6829145,0.2335784,0.04406772,0.006224064,0.03033756,0.002877746],"domain_scores_gemma":[0.4663816,0.2941226,0.02684109,0.04487038,0.1643016,0.003482698],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0005700478,0.002063295,0.03165243,0.01154775,0.0004091481,0.001006765,0.08177026,0.01239414,0.01298182,0.04267985,0.02857756,0.774347],"study_design_scores_gemma":[0.001815837,0.004507876,0.09956447,0.0581764,0.001558105,0.002839594,0.08040565,0.1013293,0.04878927,0.2118069,0.3877248,0.001481792],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.09593164,0.002171603,0.8270212,0.005566353,0.0006533944,0.04707135,0.0008961101,0.001061521,0.01962689],"genre_scores_gemma":[0.08441492,0.0005440603,0.8949152,0.0005726355,0.00003623585,0.01792719,0.0006310337,0.0001285536,0.0008302553],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.2735832,"threshold_uncertainty_score":0.8958012,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8167830120164873,"score_gpt":0.6580040613090576,"score_spread":0.1587789507074296,"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."}}