{"id":"W7107871347","doi":"10.1016/j.ufug.2025.129192","title":"Variability and bias in likelihood of urban tree failure ratings","year":2025,"lang":"en","type":"article","venue":"Urban forestry & urban greening","topic":"Tree Root and Stability Studies","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Credential; Tree (set theory); Risk perception; Perception; Test (biology); Risk assessment","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0005042003,0.0003488451,0.0005878208,0.0002558949,0.0001027827,0.00004099463,0.0002368713,0.0002015382,0.00001550783],"category_scores_gemma":[0.000363768,0.0003525255,0.0001238477,0.0006317514,0.0001820702,0.0002057456,0.000157779,0.0004417759,0.000002029983],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001028396,"about_ca_system_score_gemma":0.00004632116,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001600996,"about_ca_topic_score_gemma":0.004533546,"domain_scores_codex":[0.9981073,0.00007827025,0.0006747267,0.0004340488,0.0002027288,0.0005029215],"domain_scores_gemma":[0.9988754,0.0004184799,0.00008038909,0.0004655266,0.00006521489,0.00009495043],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002626096,0.00003998725,0.9862209,0.00039692,0.00008194814,0.000005648116,0.001817187,0.0002207458,0.0003227443,0.001114383,0.006670684,0.003082598],"study_design_scores_gemma":[0.001043018,0.00007901793,0.9828812,0.0003487763,0.00006735815,0.000002474929,0.0004952014,0.008451883,0.0006797907,0.00173701,0.003827206,0.0003870119],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9916367,0.001814937,0.001109659,0.0001712008,0.0001951515,0.0003164423,0.00002393648,0.0002496616,0.004482295],"genre_scores_gemma":[0.9983096,0.000008560244,0.001258728,0.0000139766,0.00008974776,0.0000346435,0.0000104709,0.00003173431,0.0002425065],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008231137,"threshold_uncertainty_score":0.9998927,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01119764193499114,"score_gpt":0.2110399543760306,"score_spread":0.1998423124410394,"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."}}