{"id":"W4402729535","doi":"10.1007/s00216-024-05394-7","title":"Reference material stability challenge","year":2024,"lang":"en","type":"article","venue":"Analytical and Bioanalytical Chemistry","topic":"Civil and Structural Engineering Research","field":"Engineering","cited_by":3,"is_retracted":false,"has_abstract":false,"ca_institutions":"National Research Council Canada","funders":"","keywords":"Medical laboratory; Engineering; Analytical Chemistry (journal); Environmental chemistry; Chemistry; Medicine; Pathology","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":[0.0001058234,0.000177199,0.0002026952,0.00002870367,0.0000291423,0.0001406734,0.0001119,0.0001482267,0.001821651],"category_scores_gemma":[0.00003994764,0.0001336017,0.00006792541,0.0001878641,0.0001636073,0.0000691427,0.00006060384,0.0003501227,0.00004400196],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004680463,"about_ca_system_score_gemma":0.00001618176,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000007467143,"about_ca_topic_score_gemma":0.000001661313,"domain_scores_codex":[0.9989604,0.000005194512,0.0001982181,0.0002911401,0.0002065346,0.0003385654],"domain_scores_gemma":[0.9994574,0.00008590145,0.000002848083,0.000169925,0.00002368816,0.0002602176],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000322707,0.0004103909,0.001537958,0.03513392,0.003066187,0.002156585,0.0004707732,0.00081259,0.4964673,0.1295346,0.01540986,0.3146771],"study_design_scores_gemma":[0.0001851928,0.00005395365,0.0008275681,0.0001326402,0.0001306824,0.00007635711,0.00004926993,0.9172526,0.04311056,0.004251633,0.03323704,0.0006924936],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9201741,0.001945464,0.0003067199,0.00103334,0.0001943374,0.00007919176,0.0001211419,0.0009368235,0.07520888],"genre_scores_gemma":[0.9989479,0.000286729,0.0000365985,0.000006452758,0.0002142483,0.000003977937,0.00001946696,0.00001681331,0.0004678383],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.91644,"threshold_uncertainty_score":0.9990908,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02161394056508214,"score_gpt":0.2446883555732232,"score_spread":0.2230744150081411,"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."}}