{"id":"W6931197181","doi":"10.5281/zenodo.4010218","title":"Conclusion Stability for Natural Language Based Mining of Design Discussions: Replication Package","year":2020,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Fetal and Pediatric Neurological Disorders","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Victoria","funders":"","keywords":"Stability (learning theory); R package; Natural language; Replication (statistics); Natural (archaeology)","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01793445,0.002806775,0.001398228,0.002253959,0.001158192,0.002752695,0.004050306,0.001477638,0.1683892],"category_scores_gemma":[0.08586783,0.002245107,0.003522222,0.001446407,0.0008236938,0.00368703,0.00261528,0.003308463,0.06247338],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00124116,"about_ca_system_score_gemma":0.003093097,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002948887,"about_ca_topic_score_gemma":0.002936879,"domain_scores_codex":[0.989943,0.003686566,0.0007470169,0.002614503,0.002625275,0.0003836163],"domain_scores_gemma":[0.9252598,0.04581924,0.001968052,0.01772161,0.008684106,0.0005472206],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00182223,0.0005261102,0.007208218,0.002123792,0.0006163524,0.0001609252,0.0006664364,0.01412041,0.004847226,0.01574138,0.7269145,0.2252524],"study_design_scores_gemma":[0.002713697,0.001202717,0.01225799,0.0006876094,0.0007043154,0.0004746512,0.0005887833,0.4707258,0.04454373,0.1025231,0.3631462,0.0004314383],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"dataset","genre_scores_codex":[0.003253894,0.00007074216,0.6572919,0.000500146,0.000304198,0.0007475144,0.04748196,0.2832885,0.007061191],"genre_scores_gemma":[0.07655761,0.0001146005,0.7485713,0.0003318334,0.0002385468,0.007954247,0.07811937,0.07554016,0.01257242],"genre_candidate":"dataset","genre_consensus":null,"teacher_disagreement_score":0.9820656,"threshold_uncertainty_score":0.5633181,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05321365043043509,"score_gpt":0.2813303507309438,"score_spread":0.2281167003005087,"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."}}