{"id":"W3111004241","doi":"10.1371/journal.pone.0248424","title":"On measuring agreement with numerically bounded linguistic probability schemes: A re-analysis of data from Wintle, Fraser, Wills, Nicholson, and Fidler (2019)","year":2021,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Misinformation and Its Impacts","field":"Social Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"Defence Research and Development Canada","funders":"","keywords":"Range (aeronautics); Measure (data warehouse); Term (time); Probabilistic logic; Statistics; Bounded function; Mathematics; Equivalence (formal languages); Set (abstract data type); Binomial distribution; Computer science; Linguistics; Econometrics; Discrete mathematics; Data mining; Mathematical analysis; Physics","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0006775315,0.0001011732,0.0003385579,0.00007025879,0.0001706234,0.0001137499,0.0002547293,0.00005448207,0.001219421],"category_scores_gemma":[0.004155333,0.00008628028,0.00003386267,0.0006493017,0.0001359238,0.0002703007,0.0001070652,0.0001257842,0.00001905715],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008034517,"about_ca_system_score_gemma":0.0003340647,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001062492,"about_ca_topic_score_gemma":0.005122378,"domain_scores_codex":[0.9981526,0.0001327042,0.0003095891,0.0003007136,0.000902753,0.0002016184],"domain_scores_gemma":[0.9983814,0.0002127665,0.0001801438,0.0006866851,0.0003749898,0.0001639896],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002620375,0.04691233,0.4014827,0.002429406,0.08463887,0.00009735047,0.3543701,0.0006374048,0.01573797,0.03759,0.02102293,0.03246055],"study_design_scores_gemma":[0.01821425,0.002742557,0.6297749,0.008972578,0.0463659,9.134802e-7,0.0589486,0.06536882,0.0615901,0.03903699,0.0625126,0.006471751],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9867151,0.0001673367,0.0003895691,0.0007727991,0.0000172517,0.0002252163,0.0001392329,0.00002897993,0.01154447],"genre_scores_gemma":[0.9910781,0.00010771,0.007660154,0.0003952489,0.0000438899,0.000002449016,0.000191984,0.000006181864,0.0005142387],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2954215,"threshold_uncertainty_score":0.9996936,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.175525296558567,"score_gpt":0.3166627963450355,"score_spread":0.1411374997864684,"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."}}