{"id":"W4413452402","doi":"10.31234/osf.io/rqexa_v1","title":"Think outside the box: Making up causal hypotheses from unreliable evidence","year":2025,"lang":"en","type":"article","venue":"","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"Dalhousie University","keywords":"Psychology; Econometrics; Data science; Epistemology; Computer science; Economics; Philosophy","routes":{"ca_aff":true,"ca_fund":true,"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":[0.01031284,0.0006881338,0.0009615464,0.002147368,0.0008353214,0.004091945,0.003002781,0.001771643,0.003166161],"category_scores_gemma":[0.08699891,0.001362534,0.001836627,0.001302718,0.006208555,0.008331019,0.002659469,0.002569875,0.0003447252],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001219982,"about_ca_system_score_gemma":0.001802296,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003290628,"about_ca_topic_score_gemma":0.003003552,"domain_scores_codex":[0.9950767,0.002863408,0.0002539098,0.0008076182,0.0007580369,0.0002404759],"domain_scores_gemma":[0.9160773,0.07351863,0.004345208,0.003937061,0.001344894,0.0007770113],"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":[0.0006393074,0.0001975061,0.03532879,0.0006382593,0.0005304752,0.001526927,0.005069342,0.2297544,0.004714018,0.5733599,0.001925775,0.1463152],"study_design_scores_gemma":[0.00007653884,0.0000575985,0.00260657,0.00008865952,0.00009062467,0.0002892484,0.0003488875,0.4223847,0.003010489,0.5697538,0.001244818,0.00004805976],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2067047,0.0003885973,0.7858826,0.002400436,0.00003260212,0.0001137876,0.0001809471,0.000361962,0.003934394],"genre_scores_gemma":[0.7722321,0.0002771406,0.2261096,0.0002537858,0.00004593668,0.0001162891,0.0001754325,0.00008381283,0.0007058936],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9896871,"threshold_uncertainty_score":0.05454016,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2685693378413435,"score_gpt":0.4539863764565206,"score_spread":0.1854170386151772,"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."}}