{"id":"W4377832494","doi":"10.1257/pandp.20231009","title":"Learning Curve: Progress in the Replication Crisis","year":2023,"lang":"en","type":"article","venue":"AEA Papers and Proceedings","topic":"School Choice and Performance","field":"Social Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Impact","funders":"","keywords":"Replication (statistics); Fidelity; Scale (ratio); Computer science; Learning curve; Psychology; Geography; Mathematics; Statistics; Telecommunications; Cartography","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":[],"consensus_categories":[],"category_scores_codex":[0.001291103,0.00004436931,0.00004880567,0.00004292602,0.0004500932,0.0001337787,0.000129873,0.00004533688,0.0000200296],"category_scores_gemma":[0.0001221865,0.00003201964,0.00001430698,0.0006292642,0.00006670572,0.000215835,0.00001592937,0.0001338947,0.00002547797],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001007076,"about_ca_system_score_gemma":0.00001552213,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002334868,"about_ca_topic_score_gemma":0.000133089,"domain_scores_codex":[0.9993931,0.00001641098,0.00007405992,0.0001598789,0.0001801387,0.0001764222],"domain_scores_gemma":[0.9998121,0.00004437304,0.00003479727,0.00004458402,0.00003439459,0.00002978542],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000005638066,0.000004979868,0.8957763,0.00001516436,0.000001955579,5.823675e-7,0.05968648,3.629355e-7,0.00004114096,0.002695799,0.009921518,0.03185013],"study_design_scores_gemma":[0.00009036948,0.00003435988,0.5442519,0.00001703229,0.000004025822,8.918761e-7,0.05760334,0.00005116589,0.00001779944,0.0004083628,0.397446,0.00007479079],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9466039,0.0001826352,3.759527e-8,0.01736293,0.00003142588,0.0001090102,2.208329e-7,0.00007051733,0.03563931],"genre_scores_gemma":[0.9974175,0.001176014,0.00001186182,0.0004062722,0.0001886275,0.00005244349,0.000002056132,0.000003902714,0.0007413334],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3875245,"threshold_uncertainty_score":0.3461798,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02538951414668429,"score_gpt":0.32865587355646,"score_spread":0.3032663594097758,"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."}}