{"id":"W4389921041","doi":"10.1080/00401706.2023.2296465","title":"Assessing Measurement System Agreement in the Presence of Reproducibility and Repeatability","year":2023,"lang":"en","type":"article","venue":"Technometrics","topic":"Scientific Measurement and Uncertainty Evaluation","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Repeatability; Reproducibility; Replicate; Context (archaeology); Metric (unit); System of measurement; Statistics; Computer science; Measurement uncertainty; Observational error; Accuracy and precision; Mathematics; Engineering; Physics; Geography","routes":{"ca_aff":true,"ca_fund":true,"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"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2089807,0.00009373212,0.0002303875,0.001407764,0.0001416457,0.0002682615,0.0008576486,0.00005029629,0.00001941998],"category_scores_gemma":[0.1039335,0.00005631821,0.00005293076,0.01540475,0.0001551287,0.0003081705,0.000211028,0.0001182578,0.00002179959],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001977619,"about_ca_system_score_gemma":0.00008664218,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007557331,"about_ca_topic_score_gemma":0.0000564979,"domain_scores_codex":[0.988383,0.0007868854,0.001310408,0.001455453,0.007778253,0.0002859761],"domain_scores_gemma":[0.9939989,0.001573135,0.0003700405,0.003018345,0.001002818,0.00003669271],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001904345,0.0001868634,0.7192077,0.000134233,0.000008786311,0.000003561317,0.001771859,0.0003601123,0.00425052,0.002358095,0.00265311,0.2690461],"study_design_scores_gemma":[0.0002489988,0.00004632158,0.9593733,0.00006457002,0.00001340451,0.000001506349,0.01371259,0.01405578,0.002374422,0.009330564,0.0006722028,0.0001063271],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9932387,0.0002902996,0.002477197,0.0008068945,0.0003235655,0.000733018,0.00000387476,0.00007044026,0.002056017],"genre_scores_gemma":[0.9995629,0.000006410995,0.0003220564,0.000008250357,0.00001487512,0.00003079173,0.000001628443,0.000002648314,0.00005043219],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2689398,"threshold_uncertainty_score":0.9036144,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.590017664249239,"score_gpt":0.4596789235328383,"score_spread":0.1303387407164007,"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."}}