{"id":"W7082293784","doi":"10.48448/4a71-5066","title":"LLMs can Perform Multi-Dimensional Analytic Writing Assessments: A Case Study of L2 Graduate-Level Academic English Writing","year":2025,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"Geochemistry and Geologic Mapping","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Saskatchewan","funders":"","keywords":"Context (archaeology); Quality (philosophy); Task (project management); Code (set theory); Academic writing; Graduate students","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":[0.02586432,0.0009110394,0.0006068127,0.004098144,0.002202354,0.002691547,0.001648694,0.001599671,0.003646091],"category_scores_gemma":[0.1666538,0.0004253352,0.0005093132,0.003379343,0.001327755,0.003385665,0.004924737,0.001661862,0.002985331],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001483769,"about_ca_system_score_gemma":0.002193019,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003001422,"about_ca_topic_score_gemma":0.007186133,"domain_scores_codex":[0.9748906,0.01801358,0.001932686,0.001381572,0.003241536,0.000540121],"domain_scores_gemma":[0.7520643,0.1871641,0.009658832,0.0175263,0.02952447,0.004061928],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001159646,0.001864466,0.06845959,0.003080178,0.0001592048,0.006465832,0.2183041,0.005170746,0.03019605,0.00525098,0.05606736,0.6038218],"study_design_scores_gemma":[0.0009377142,0.004038008,0.1457476,0.002244776,0.0002749617,0.007583664,0.247433,0.1286923,0.08605204,0.02509201,0.3507085,0.001195373],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9246503,0.0006709242,0.04709066,0.002935308,0.0002163345,0.0009802686,0.002006451,0.006278546,0.01517111],"genre_scores_gemma":[0.8589522,0.0002792485,0.1302626,0.0008378418,0.0001248363,0.001063328,0.001578646,0.00104658,0.005854704],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02586432,"threshold_uncertainty_score":0.1367852,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1016390655937194,"score_gpt":0.3582165939705774,"score_spread":0.256577528376858,"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."}}