{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.4451979,0.001519517,0.003126564,0.01004474,0.005265606,0.02175447,0.004440198,0.005472002,0.005156337],"category_scores_gemma":[0.5600744,0.0009180261,0.003291928,0.006430899,0.0174457,0.02317963,0.01569265,0.008436005,0.0006867903],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02815798,"about_ca_system_score_gemma":0.0492145,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005034302,"about_ca_topic_score_gemma":0.0092139,"domain_scores_codex":[0.3311875,0.6152956,0.02018051,0.004247623,0.02581914,0.003269612],"domain_scores_gemma":[0.2658109,0.6123998,0.03384253,0.03202028,0.04734644,0.008580032],"domain_codex":"methods","domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.001529446,0.001656088,0.0301463,0.04213761,0.002323651,0.0005844028,0.08378033,0.003304191,0.001404217,0.1415842,0.009689775,0.6818598],"study_design_scores_gemma":[0.001931821,0.01321384,0.06718425,0.2829574,0.004225684,0.001379331,0.1654906,0.01705247,0.009210953,0.318053,0.1184305,0.0008701141],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2238963,0.08172827,0.3297276,0.2635749,0.003913144,0.02209442,0.000679584,0.000560709,0.07382514],"genre_scores_gemma":[0.7144811,0.008655936,0.258077,0.008000887,0.0002631808,0.009430978,0.0001085862,0.0001015059,0.0008807761],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4451979,"threshold_uncertainty_score":0.6841698,"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."}}