{"id":"W7117704533","doi":"10.18552/joaw.v15i2.1346","title":"Student Evaluative Judgements of Writing and Artificial Intelligence: The Disconnect between Structural and Conceptual Knowledge","year":2025,"lang":"","type":"article","venue":"Journal of Academic Writing","topic":"Writing and Handwriting Education","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Judgement; Generative grammar; Quality (philosophy); Professional writing; Writing assessment; Work (physics); Implicit knowledge","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":["sts"],"consensus_categories":[],"category_scores_codex":[0.01126682,0.0002955932,0.0007532344,0.0003135175,0.001459839,0.0002768906,0.0006250791,0.000330728,0.00003792788],"category_scores_gemma":[0.003461822,0.0002400516,0.0001592875,0.000694447,0.002006762,0.0005464897,0.0003412438,0.001907925,0.000001097426],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002036305,"about_ca_system_score_gemma":0.0007477714,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000601116,"about_ca_topic_score_gemma":0.00002020238,"domain_scores_codex":[0.9943503,0.001382014,0.002494044,0.0003635932,0.0008565004,0.0005535556],"domain_scores_gemma":[0.9921392,0.004611754,0.002011135,0.0001265579,0.0009251363,0.0001862549],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.00007754684,0.00008100721,0.2666139,0.0003740152,0.0005661655,0.000003001182,0.2442274,0.00003153353,0.001340722,0.07427289,0.000245456,0.4121664],"study_design_scores_gemma":[0.0005110072,0.0002986415,0.107504,0.005102873,0.0007950431,0.00001807883,0.8595145,0.0005072257,0.001967643,0.02328164,0.0001839231,0.0003154398],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9749273,0.01656967,0.0002384944,0.004922005,0.0008056731,0.0003895868,0.00001671948,0.00000919839,0.002121374],"genre_scores_gemma":[0.9945764,0.001983346,0.0002466306,0.00009294068,0.002982596,0.00000430072,0.000001485137,0.00001517886,0.00009710609],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6152871,"threshold_uncertainty_score":0.9998401,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0794952975947915,"score_gpt":0.453795789149932,"score_spread":0.3743004915551404,"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."}}