{"id":"W4415396057","doi":"10.1080/15391523.2025.2568526","title":"Keeping the “glass box” transparent: Comparing expert and AI-generated ratings and feedback in stealth assessment for judgement-focused negotiation simulations","year":2025,"lang":"en","type":"article","venue":"Journal of Research on Technology in Education","topic":"Conflict Management and Negotiation","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Negotiation; Electronic learning; Online assessment; Formative assessment; Higher education; Work (physics)","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":[],"consensus_categories":[],"category_scores_codex":[0.002672663,0.00005845253,0.0001350558,0.001431199,0.0004422197,0.0001143062,0.0001649,0.0000916522,0.000004631811],"category_scores_gemma":[0.0005289591,0.00005155913,0.00001556985,0.001195794,0.0001493852,0.000242043,0.00002820902,0.0003806186,1.712802e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004307341,"about_ca_system_score_gemma":0.0004878242,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003825476,"about_ca_topic_score_gemma":0.003759774,"domain_scores_codex":[0.9987428,0.0003006095,0.0003494104,0.0001340583,0.0002669891,0.000206189],"domain_scores_gemma":[0.9989228,0.0004834151,0.0001537608,0.00008763666,0.0003248429,0.00002755123],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002064026,0.0007202575,0.3917575,0.0001287978,0.00006610145,8.918977e-7,0.02226666,0.001931253,0.003212346,0.2380147,0.002366828,0.3393282],"study_design_scores_gemma":[0.005841194,0.0007443421,0.6847707,0.001561383,0.00004827391,0.000001165561,0.1312123,0.05605429,0.002415093,0.07824936,0.03877629,0.0003256103],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8864547,0.0004739603,0.001899974,0.1095895,0.0002334678,0.0007961278,4.772835e-7,0.00001219739,0.0005396638],"genre_scores_gemma":[0.998237,0.0007453064,0.0005506075,0.0002250751,0.00006844854,0.00005691475,0.000003209363,0.000003739043,0.0001097645],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3390026,"threshold_uncertainty_score":0.3401241,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1056071517416213,"score_gpt":0.498568736867848,"score_spread":0.3929615851262266,"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."}}