{"id":"W2307814545","doi":"10.1007/978-3-319-30671-1_30","title":"Toward Reproducible Baselines: The Open-Source IR Reproducibility Challenge","year":2016,"lang":"en","type":"book-chapter","venue":"Lecture notes in computer science","topic":"Information Retrieval and Search Behavior","field":"Computer Science","cited_by":90,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Computer science; Open source; Information retrieval; World Wide Web; Source code; Data science; Software; Programming language","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":["metaepi_narrow","scholarly_communication","open_science"],"consensus_categories":[],"category_scores_codex":[0.00925406,0.0005629864,0.0005872521,0.000468231,0.0006127947,0.001401116,0.01121854,0.0002877162,0.0001165246],"category_scores_gemma":[0.0009958389,0.0003301203,0.0001797069,0.0008398675,0.001018158,0.001751373,0.006752414,0.0009977368,0.0001818056],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003156411,"about_ca_system_score_gemma":0.001070555,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000537423,"about_ca_topic_score_gemma":0.00003058507,"domain_scores_codex":[0.9926118,0.000119481,0.0009166752,0.003910317,0.001587539,0.0008541927],"domain_scores_gemma":[0.9898199,0.0006356,0.0004540861,0.008087706,0.0007831211,0.0002195175],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00001361792,0.00004657966,0.00004317526,0.00004271083,0.000008162523,0.00002300424,0.001760199,0.0009536639,0.00004833169,0.02367813,0.0001373715,0.9732451],"study_design_scores_gemma":[0.001910315,0.001137104,0.0009977854,0.002147285,0.00004826574,0.0003301973,0.00000401483,0.2636248,0.01319574,0.4446822,0.2679442,0.003978116],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.00007408601,0.000567008,0.9634353,0.02325216,0.001987444,0.001235522,0.00001271832,0.0002290212,0.009206723],"genre_scores_gemma":[0.4285395,0.002290525,0.4772457,0.02837297,0.009203536,0.0005345094,0.00003517353,0.0003488691,0.05342925],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.969267,"threshold_uncertainty_score":0.9999151,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06459347854591166,"score_gpt":0.2974063553987016,"score_spread":0.23281287685279,"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."}}