{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1063875,0.003584481,0.004279277,0.005385949,0.003785686,0.01397727,0.01257727,0.007196176,0.007318506],"category_scores_gemma":[0.3187569,0.002293064,0.003192501,0.006255355,0.004920019,0.01725592,0.01613441,0.01039723,0.01302155],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002132603,"about_ca_system_score_gemma":0.005993261,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002856187,"about_ca_topic_score_gemma":0.003897526,"domain_scores_codex":[0.8577686,0.07060713,0.01021457,0.01765573,0.04077088,0.002983083],"domain_scores_gemma":[0.6427979,0.1269802,0.0109349,0.1719242,0.04350978,0.003853039],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.002173782,0.001195657,0.0075527,0.005294164,0.001561289,0.0004989238,0.00173495,0.01397282,0.03391204,0.03841288,0.2119023,0.6817885],"study_design_scores_gemma":[0.00142699,0.001618303,0.01309377,0.001742296,0.001116278,0.002924457,0.001861702,0.2908488,0.09789767,0.36045,0.2263255,0.0006942771],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04566764,0.02140775,0.7979408,0.01204104,0.005706161,0.001135703,0.01307855,0.08376784,0.01925453],"genre_scores_gemma":[0.2331636,0.00399811,0.6690294,0.005533408,0.003308087,0.001466864,0.04603302,0.02549885,0.01196867],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8936125,"threshold_uncertainty_score":0.5626375,"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."}}