{"id":"W4404690134","doi":"10.4102/aej.v12i1.758","title":"A rocky road to evidence: Evaluating literacy programmes using a trust-based approach in a context of fragility","year":2024,"lang":"en","type":"article","venue":"African Evaluation Journal","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Context (archaeology); Computer science; Data collection; Fragility; Quality (philosophy); Baseline (sea); Data science; Knowledge management; Political science","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","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05709198,0.0002261046,0.0004689162,0.001453812,0.000217884,0.001011139,0.0006221046,0.00008510676,0.00154312],"category_scores_gemma":[0.01073472,0.0001683402,0.0002357189,0.003627766,0.00008132522,0.001458886,0.00007911719,0.0004671852,0.00006195584],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006843295,"about_ca_system_score_gemma":0.00401098,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006312559,"about_ca_topic_score_gemma":0.00003900517,"domain_scores_codex":[0.9880577,0.002588638,0.002032772,0.0006215371,0.006292139,0.000407204],"domain_scores_gemma":[0.9939734,0.001490943,0.0006281564,0.0004991202,0.003156674,0.0002516446],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001663027,0.0001348901,0.0275375,0.000035496,0.0000261724,0.00000337248,0.01808261,0.08217331,0.0008668068,0.00007948736,0.0002114647,0.8706826],"study_design_scores_gemma":[0.0007729633,0.0004101442,0.03448946,0.0005218298,0.00007404631,0.00003353973,0.009231961,0.9509021,0.0001813656,0.002444317,0.0007553648,0.0001829073],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9457781,0.00178244,0.0478321,0.002287688,0.0004617792,0.00117567,0.00000809087,0.00002627056,0.0006478071],"genre_scores_gemma":[0.9773336,0.00000702179,0.02195561,0.0002850019,0.0001950253,0.0001241882,0.000004466263,0.00001623792,0.00007881736],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8704997,"threshold_uncertainty_score":0.9993696,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4099295910482118,"score_gpt":0.576455307169125,"score_spread":0.1665257161209132,"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."}}