{"id":"W2777631817","doi":"10.1177/2515245918787489","title":"A Unified Framework to Quantify the Credibility of Scientific Findings","year":2018,"lang":"en","type":"article","venue":"Advances in Methods and Practices in Psychological Science","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":229,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"","keywords":"Credibility; Scrutiny; Workflow; Data science; Transparency (behavior); Computer science; Falsifiability; Robustness (evolution); Crowdsourcing; Management science; Psychology; Epistemology; Political science; World Wide Web; Economics","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":["metaresearch"],"category_scores_codex":[0.527731,0.006605445,0.007486825,0.06738781,0.005925435,0.0275579,0.01302755,0.01297281,0.005375884],"category_scores_gemma":[0.7295894,0.00325452,0.01248216,0.02949438,0.02631092,0.02644787,0.02231801,0.01378249,0.001556185],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.015411,"about_ca_system_score_gemma":0.02533516,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01068896,"about_ca_topic_score_gemma":0.004788409,"domain_scores_codex":[0.4417467,0.4069107,0.05615048,0.02857879,0.06306949,0.003543927],"domain_scores_gemma":[0.2094469,0.6218076,0.05745832,0.06076817,0.04801971,0.002499256],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002453804,0.0001352306,0.008009824,0.004392453,0.002784602,0.0002523281,0.003741271,0.02709891,0.0008553779,0.8264104,0.007363636,0.1187106],"study_design_scores_gemma":[0.0001569254,0.0002446547,0.003417641,0.003057608,0.001028188,0.0003435324,0.0006230418,0.03613265,0.001042347,0.9362923,0.01731522,0.0003458216],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.003079564,0.006102644,0.97288,0.007454493,0.0004700053,0.002125036,0.00123067,0.0005609669,0.00609669],"genre_scores_gemma":[0.08710752,0.002544815,0.9002799,0.001495044,0.0007425328,0.006466692,0.000743287,0.0001704208,0.0004497099],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.472269,"threshold_uncertainty_score":0.5823918,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7627076276557118,"score_gpt":0.7394797943067195,"score_spread":0.02322783334899226,"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."}}