{"id":"W7116985352","doi":"10.64898/2025.12.19.695552","title":"Petal to the metal: The slow road to automating large-scale phenology labeling for herbarium specimens","year":2025,"lang":"","type":"article","venue":"bioRxiv (Cold Spring Harbor Laboratory)","topic":"Remote Sensing in Agriculture","field":"Environmental Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"National Science Foundation","keywords":"Phenology; Herbarium; Digitization; Metadata; Annotation; Resource (disambiguation)","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":[],"consensus_categories":[],"category_scores_codex":[0.004493087,0.0008936266,0.001061613,0.002474517,0.001191319,0.00279792,0.002620441,0.001160899,0.008534796],"category_scores_gemma":[0.009017723,0.0007599939,0.001195144,0.002327555,0.0006891139,0.003211252,0.003454037,0.002537353,0.01031956],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001061941,"about_ca_system_score_gemma":0.001396477,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007576882,"about_ca_topic_score_gemma":0.02200373,"domain_scores_codex":[0.9977412,0.0004576287,0.0001397292,0.0008274845,0.0006858868,0.0001480146],"domain_scores_gemma":[0.9897709,0.00207045,0.0006323532,0.005029731,0.001885846,0.0006108588],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00052634,0.0001772266,0.04073307,0.001143681,0.0004301021,0.0002310784,0.0008013515,0.004477065,0.07375102,0.004207494,0.3410405,0.5324811],"study_design_scores_gemma":[0.0001179779,0.0002133291,0.08768373,0.0006841241,0.0001639337,0.0007303155,0.001356438,0.07817054,0.0628638,0.02269491,0.7451329,0.0001880063],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1529903,0.006382028,0.5266804,0.009867691,0.00185088,0.0006385702,0.1163844,0.1623437,0.02286205],"genre_scores_gemma":[0.1029353,0.00144241,0.7706546,0.002732581,0.0003273906,0.000598253,0.1042651,0.00845394,0.008590441],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.008534796,"threshold_uncertainty_score":0.0285517,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.007400851747412249,"score_gpt":0.2147454323536735,"score_spread":0.2073445806062612,"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."}}