{"id":"W4396927620","doi":"10.1515/cclm-2024-0527","title":"The final part of the CRESS trilogy – how to evaluate the quality of stability studies","year":2024,"lang":"en","type":"article","venue":"Clinical Chemistry and Laboratory Medicine (CCLM)","topic":"Clinical Laboratory Practices and Quality Control","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; Hôpital Maisonneuve-Rosemont","funders":"","keywords":"Checklist; Quality (philosophy); Computer science; Sample (material); Stability (learning theory); Resource (disambiguation); Risk analysis (engineering); Plan (archaeology); Reliability engineering; Data mining; Engineering; Machine learning; Medicine; Psychology; Chemistry","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":["metaresearch"],"category_scores_codex":[0.3040338,0.002265042,0.003810472,0.0194307,0.004632853,0.01596338,0.005465909,0.006097747,0.02086863],"category_scores_gemma":[0.6248518,0.002371843,0.006106333,0.007550552,0.007232698,0.01080827,0.008003891,0.008599052,0.02153724],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.009062117,"about_ca_system_score_gemma":0.05583782,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003472665,"about_ca_topic_score_gemma":0.005640004,"domain_scores_codex":[0.6617962,0.1741686,0.09526246,0.005304375,0.0604821,0.002986276],"domain_scores_gemma":[0.2260208,0.2921571,0.08360348,0.05497132,0.3289103,0.01433687],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0005073248,0.0001578346,0.003548475,0.02138002,0.000434313,0.0003296573,0.004065755,0.0003698639,0.001455831,0.007641354,0.5593385,0.400771],"study_design_scores_gemma":[0.0002202231,0.0004513822,0.006033854,0.03606361,0.0004162005,0.0008298642,0.001954972,0.0006260928,0.002251873,0.01603287,0.9347824,0.0003367036],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"methods","genre_scores_codex":[0.008978155,0.07273646,0.2085899,0.4554662,0.07607763,0.08016328,0.01657191,0.0123928,0.06902366],"genre_scores_gemma":[0.03120435,0.04051486,0.7764923,0.0448732,0.01578651,0.0509569,0.007662727,0.004226112,0.028283],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6959662,"threshold_uncertainty_score":0.8582502,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3497082966979214,"score_gpt":0.5294942181390766,"score_spread":0.1797859214411552,"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."}}