{"id":"W2287148974","doi":"10.1145/2888422.2888439","title":"Report on the SIGIR 2015 Workshop on Reproducibility, Inexplicability, and Generalizability of Results (RIGOR)","year":2016,"lang":"en","type":"article","venue":"ACM SIGIR Forum","topic":"Web Data Mining and Analysis","field":"Computer Science","cited_by":59,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"Università degli Studi di Padova","keywords":"Generalizability theory; Computer science; Presentation (obstetrics); Thursday; Information retrieval; Library science; Data science; Statistics; Medicine","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.005688907,0.0002303333,0.0003563738,0.0000948482,0.0001641046,0.00008632347,0.001833374,0.0001002398,0.00001391268],"category_scores_gemma":[0.01669456,0.0001195423,0.0001371179,0.0005233134,0.0002956594,0.000311224,0.001604173,0.0001604857,0.00002898022],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005641903,"about_ca_system_score_gemma":0.00007158014,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001071751,"about_ca_topic_score_gemma":0.00004342706,"domain_scores_codex":[0.9960001,0.0002854286,0.0007569217,0.002041582,0.0005220588,0.0003939852],"domain_scores_gemma":[0.9848558,0.001966414,0.0003536961,0.01252153,0.0001779223,0.0001246765],"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":[0.0006073102,0.00169714,0.1901527,0.00007912196,0.0003186968,0.0001051576,0.001250064,0.0001737375,0.01034418,0.05430614,0.3245273,0.4164384],"study_design_scores_gemma":[0.005299484,0.003164568,0.3087355,0.001043032,0.0002228294,0.0002348182,0.0005120747,0.005977084,0.1449754,0.3671139,0.159782,0.002939396],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.849148,0.00007002734,0.004623293,0.143811,0.0002329134,0.000299111,0.00007167718,0.0001466699,0.001597285],"genre_scores_gemma":[0.9955367,0.00002904775,0.002626835,0.0007002136,0.00008035628,0.00002301538,0.00001066381,0.00001120439,0.0009819471],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.413499,"threshold_uncertainty_score":0.9915882,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04259519423738254,"score_gpt":0.3006233049999692,"score_spread":0.2580281107625866,"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."}}