{"id":"W4200412695","doi":"10.1111/jep.13646","title":"Monitoring the integrity and usability of policy evaluation tools within an evolving sociocultural context: A demonstration of reflexivity using the CFPC Family Medicine Longitudinal Survey","year":2021,"lang":"en","type":"article","venue":"Journal of Evaluation in Clinical Practice","topic":"Primary Care and Health Outcomes","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Alberta Advanced Education; Alberta Medical Association; University of Alberta","funders":"","keywords":"Reflexivity; Context (archaeology); Usability; Sociocultural evolution; Medicine; Medical education; Psychology; Nursing; Sociology; Computer science","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.6824104,0.0009285645,0.001221705,0.005356836,0.005098969,0.009838765,0.003098974,0.001751154,0.0008879679],"category_scores_gemma":[0.7622214,0.001113645,0.001541563,0.004185766,0.01041403,0.008159925,0.00894466,0.003416561,0.0002629596],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01232505,"about_ca_system_score_gemma":0.02787743,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009101355,"about_ca_topic_score_gemma":0.008729912,"domain_scores_codex":[0.2453058,0.6834899,0.03142183,0.007674188,0.02877579,0.003332478],"domain_scores_gemma":[0.08990718,0.755501,0.02895239,0.06304703,0.06118591,0.001406411],"domain_codex":"methods","domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.0008943435,0.001220969,0.1392485,0.002232174,0.0005606429,0.000337458,0.5815469,0.002558406,0.003076914,0.009685748,0.003072395,0.2555656],"study_design_scores_gemma":[0.00153763,0.008620284,0.3149576,0.01266803,0.001200105,0.0009941211,0.4481076,0.04950184,0.03958616,0.04130746,0.08021761,0.001301643],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8079872,0.0006795305,0.153285,0.006377279,0.0002891525,0.01382727,0.0005198421,0.0006660517,0.01636867],"genre_scores_gemma":[0.916728,0.0001679023,0.07442281,0.0006552805,0.00004817823,0.007331827,0.0001065797,0.0001295344,0.0004098092],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6824104,"threshold_uncertainty_score":0.3916445,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6851252036797749,"score_gpt":0.6801626173874009,"score_spread":0.004962586292374005,"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."}}