{"id":"W4407559766","doi":"10.1111/spc3.70042","title":"A Credibility Revolution for Relationship Science: Where Can We Step Up Our Game?","year":2025,"lang":"en","type":"article","venue":"Social and Personality Psychology Compass","topic":"Behavioral Health and Interventions","field":"Psychology","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Credibility; Psychology; Social psychology; Epistemology; 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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001236908,0.000154194,0.0002822497,0.0001569458,0.001068205,0.00003538944,0.0002319161,0.0002939655,0.0002115486],"category_scores_gemma":[0.00007205184,0.0001585696,0.0001787397,0.0005171295,0.0006253471,0.00008650021,0.00004073948,0.0004034316,0.00002140695],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001476082,"about_ca_system_score_gemma":0.0001898827,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001082584,"about_ca_topic_score_gemma":0.001710676,"domain_scores_codex":[0.9980211,0.0003272308,0.0004054801,0.0006060303,0.0001422347,0.0004979087],"domain_scores_gemma":[0.9991363,0.00009452133,0.0001293623,0.0002622918,0.00023686,0.0001406731],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006716436,0.0006016146,0.4449531,0.0001651773,0.00005622581,0.000001150315,0.006450333,6.755854e-8,0.0000602226,0.2697299,0.2330377,0.04427281],"study_design_scores_gemma":[0.001455432,0.000144032,0.9302506,0.00003689198,0.00005150989,0.00000460854,0.003320058,0.00003501254,7.738178e-7,0.01942992,0.04513449,0.000136665],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8740437,0.001438382,0.003013519,0.1065722,0.004988925,0.0008091494,0.0003598375,0.0001075803,0.008666706],"genre_scores_gemma":[0.9914384,0.00001325751,0.0001986962,0.001303457,0.0002893909,0.0002027564,0.00003177869,0.000007852202,0.006514422],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4852975,"threshold_uncertainty_score":0.821588,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1964039716477773,"score_gpt":0.5057994198532636,"score_spread":0.3093954482054863,"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."}}