{"id":"W4319439102","doi":"10.31038/ijnm.2022332","title":"Peer-to-Peer Evaluation: Kritik Pilot at Two Post- secondary Canadian Institutions","year":2022,"lang":"en","type":"article","venue":"","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Winnipeg; Brandon University","funders":"","keywords":"Peer review; Peer-to-peer; Political science; Psychology; Medical education; Public relations; Computer science; Medicine; World Wide Web; Law","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":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002359098,0.00009016057,0.00009578391,0.0001817657,0.003582031,0.0001302735,0.000389895,0.00002263025,0.100144],"category_scores_gemma":[0.0002443465,0.000099877,0.00004956425,0.0005512859,0.0001003203,0.0002017789,0.0001886324,0.000165347,0.0007366108],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001773726,"about_ca_system_score_gemma":0.003796465,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.4706766,"about_ca_topic_score_gemma":0.9327376,"domain_scores_codex":[0.9972824,0.000266042,0.0001520984,0.0002435454,0.001652319,0.0004036274],"domain_scores_gemma":[0.9987108,0.00006012052,0.00002856152,0.0001681291,0.0006491901,0.0003832246],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0000225054,0.0001464209,0.02033015,0.000002174969,0.00005712152,0.00001583417,0.03215831,0.0002637323,0.0002366447,0.2086686,0.7275083,0.01059015],"study_design_scores_gemma":[0.0004293644,0.00009462579,0.02598516,0.000001028135,0.00002097482,0.000001671998,0.02045523,0.00003957662,0.000006130745,0.0003841379,0.9524202,0.0001618509],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.1068999,0.00004899357,0.00001301048,0.07779077,0.001379106,0.0005729811,0.00008097875,0.0000632761,0.8131509],"genre_scores_gemma":[0.8169692,0.000001698934,0.0003993879,0.004750644,0.0002416831,0.0001589889,0.000132644,0.000006941054,0.1773388],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.7100693,"threshold_uncertainty_score":0.9977152,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08764148884602259,"score_gpt":0.4014337841839782,"score_spread":0.3137922953379557,"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."}}