{"id":"W4226382178","doi":"10.48550/arxiv.2201.11308","title":"Calibration with Privacy in Peer Review","year":2022,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Access Control and Trust","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; Carnegie Mellon University; National Science Foundation","keywords":"Computer science; Adversary; Pareto principle; Block (permutation group theory); Calibration; Identity (music); Theoretical computer science; Data mining; Computer security; Information retrieval; Mathematical optimization; Mathematics; Statistics","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0006971158,0.0001455232,0.0002675124,0.0001180232,0.0002706537,0.0000574514,0.0007089331,0.0001212194,0.001436675],"category_scores_gemma":[0.0001539324,0.0001514406,0.00008560189,0.0006220476,0.0001244022,0.0003139372,0.0004895383,0.0005245035,0.00001646627],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003072922,"about_ca_system_score_gemma":0.000490908,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00505101,"about_ca_topic_score_gemma":0.003744122,"domain_scores_codex":[0.9985381,0.0003528191,0.0001344324,0.0004920745,0.0002475266,0.0002350327],"domain_scores_gemma":[0.9991544,0.00006453635,0.0001785318,0.000358198,0.0001545104,0.00008977567],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002122412,0.0003406619,0.2118079,0.0009993545,0.0001681891,0.001272248,0.005571817,0.0507542,0.000001276889,0.7183986,0.008496511,0.001976973],"study_design_scores_gemma":[0.003401456,0.000187013,0.03859437,0.002904727,0.0008153415,0.000002733719,0.006653399,0.01921556,0.000004386035,0.09274306,0.8329459,0.002531996],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4915383,0.005581851,0.01533626,0.02827405,0.00143536,0.004846755,0.0001206436,0.0006586119,0.4522081],"genre_scores_gemma":[0.9808063,0.00358193,0.00003526028,0.0003567994,0.00007391637,0.000004445473,0.00005082761,0.00001138219,0.01507912],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8244495,"threshold_uncertainty_score":0.9994761,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1088335609356486,"score_gpt":0.2441590464734186,"score_spread":0.13532548553777,"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."}}