{"id":"W6894174360","doi":"10.5281/zenodo.6762602","title":"duncster94/BIONIC-evals: v0.1.0","year":2022,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"","field":"","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002072066,0.003492021,0.001821411,0.002269907,0.001012434,0.004554019,0.005836963,0.0023495,0.4316738],"category_scores_gemma":[0.004897154,0.001772593,0.001279955,0.002076046,0.0008431495,0.003313018,0.003213675,0.00264802,0.5201613],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001912577,"about_ca_system_score_gemma":0.001289902,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007422065,"about_ca_topic_score_gemma":0.00659216,"domain_scores_codex":[0.9984331,0.0002292478,0.00006686318,0.0002457477,0.0007843388,0.0002406853],"domain_scores_gemma":[0.9979947,0.0002943556,0.0001011097,0.0007027124,0.0006058297,0.0003013234],"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":[0.0002737188,0.00002153781,0.0001226696,0.0002373672,0.00003041085,0.0000324262,0.0000354174,0.001479604,0.001148609,0.005464297,0.9712616,0.01989231],"study_design_scores_gemma":[0.0001397027,0.00003731569,0.0003063085,0.00007356573,0.00001915752,0.00009179839,0.00002291386,0.005706326,0.007723207,0.007216882,0.9785814,0.0000814693],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"software","genre_gemma":"software","genre_scores_codex":[0.001784751,0.0007937082,0.08407874,0.0008802917,0.001029359,0.0002245326,0.1331251,0.5497409,0.2283427],"genre_scores_gemma":[0.02105663,0.0005713732,0.03703065,0.0007915558,0.0003229935,0.0005902567,0.2512773,0.4466479,0.2417113],"genre_candidate":"software","genre_consensus":"software","teacher_disagreement_score":0.4316738,"threshold_uncertainty_score":0.8106483,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04115103086937642,"score_gpt":0.2575676287620931,"score_spread":0.2164165978927167,"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."}}