{"id":"W4416583604","doi":"10.2139/ssrn.5795725","title":"Forecasting Social Science: Evidence from 100 Projects","year":2025,"lang":"en","type":"preprint","venue":"SSRN Electronic Journal","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Download; Set (abstract data type); Consensus forecast; Field (mathematics); Affect (linguistics); Data set","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":[0.01279142,0.0006337056,0.0007210033,0.003838292,0.001156516,0.002350525,0.001480637,0.002848138,0.00681255],"category_scores_gemma":[0.1035083,0.0005900642,0.0007808882,0.007734836,0.001369111,0.003574082,0.002132412,0.002513333,0.001928256],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009182163,"about_ca_system_score_gemma":0.001658616,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01017153,"about_ca_topic_score_gemma":0.01194522,"domain_scores_codex":[0.9947168,0.002915136,0.0002723885,0.0005722106,0.001283423,0.0002400775],"domain_scores_gemma":[0.7686966,0.1883436,0.01781566,0.01312124,0.008193269,0.003829619],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003199395,0.002736551,0.7392671,0.0009092693,0.00138693,0.000656724,0.001629809,0.06772009,0.0003987066,0.01711738,0.02845583,0.1365222],"study_design_scores_gemma":[0.001791218,0.001850362,0.6883348,0.00086145,0.001310399,0.0004414163,0.004258839,0.1452353,0.002567586,0.1008122,0.05234877,0.0001875064],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9779826,0.002346851,0.004644621,0.002084975,0.00005438176,0.00008083995,0.005003248,0.00007798547,0.007724542],"genre_scores_gemma":[0.9847861,0.002122101,0.002631841,0.0001333321,0.00007385334,0.00009379935,0.008017176,0.0000294392,0.002112369],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9872086,"threshold_uncertainty_score":0.06764823,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2777182010040681,"score_gpt":0.4491877264042935,"score_spread":0.1714695254002254,"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."}}