{"id":"W2397669334","doi":"","title":"The Fault, Dear Researchers, is not in Cranfield, But in our Metrics, that they are Unrealistic.","year":2012,"lang":"en","type":"article","venue":"","topic":"Information Retrieval and Search Behavior","field":"Computer Science","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Computer science; Ranking (information retrieval); Measure (data warehouse); Test (biology); Data science; Information retrieval; Data mining","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","bibliometrics"],"consensus_categories":[],"category_scores_codex":[0.05673543,0.00231335,0.002381194,0.006044754,0.004557278,0.011434,0.00519648,0.005458296,0.01169485],"category_scores_gemma":[0.3862545,0.0008680344,0.00106165,0.009759136,0.01120715,0.03670719,0.006133678,0.01069875,0.009913576],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005119312,"about_ca_system_score_gemma":0.006978796,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007330072,"about_ca_topic_score_gemma":0.007699409,"domain_scores_codex":[0.89092,0.03920149,0.008999418,0.008180127,0.05124241,0.001456435],"domain_scores_gemma":[0.6340027,0.1613117,0.02759212,0.06133657,0.1099691,0.005787759],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000503372,0.0004842202,0.02937471,0.002924974,0.0003428786,0.00009293127,0.002273934,0.002707758,0.001932696,0.318937,0.320631,0.3197945],"study_design_scores_gemma":[0.0001586425,0.001352085,0.01885981,0.002736745,0.0003257093,0.000857067,0.004784329,0.01328398,0.005511888,0.5132545,0.4382749,0.0006003286],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"commentary","genre_scores_codex":[0.02526584,0.04136528,0.4778185,0.3260681,0.01428555,0.0006513578,0.004553035,0.004856474,0.1051359],"genre_scores_gemma":[0.4179718,0.01535924,0.457374,0.06296507,0.01314047,0.001680474,0.00295969,0.002698829,0.02585047],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.9939553,"threshold_uncertainty_score":0.3000493,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1519798433484288,"score_gpt":0.3618475906571991,"score_spread":0.2098677473087703,"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."}}