{"id":"W6911173347","doi":"10.5281/zenodo.10589601","title":"Paper is not enough: Crowdsourcing the T1 mapping common ground via the ISMRM reproducibility challenge","year":2024,"lang":"en","type":"dataset","venue":"PolyPublie (École Polytechnique de Montréal)","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Philips (Canada); McGill University Health Centre; University of British Columbia; McGill University; Polytechnique Montréal","funders":"","keywords":"Crowdsourcing; Common ground; Reproducibility; Task (project management); Reliability (semiconductor)","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":[0.005056677,0.005076908,0.002891218,0.003254416,0.003285829,0.003908145,0.006063223,0.004877209,0.031122],"category_scores_gemma":[0.01611032,0.001133789,0.003341058,0.003784898,0.001553088,0.002147653,0.007003092,0.003358326,0.08213226],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00215977,"about_ca_system_score_gemma":0.003654625,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.04661124,"about_ca_topic_score_gemma":0.09948006,"domain_scores_codex":[0.9942217,0.001381519,0.000338033,0.002124108,0.001372016,0.0005625965],"domain_scores_gemma":[0.9938191,0.001150542,0.000231956,0.003022084,0.001292907,0.0004834033],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002884054,0.00007945087,0.00114617,0.0005900302,0.0001602228,0.0001277444,0.00008815317,0.001570023,0.0009866302,0.0006286768,0.978124,0.01621049],"study_design_scores_gemma":[0.001014991,0.0002056058,0.009992793,0.0008461611,0.0002936148,0.0009627679,0.0006763626,0.01664185,0.007398985,0.01490239,0.9468178,0.0002467516],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.009847533,0.003666305,0.01641453,0.002246842,0.002870486,0.0005351371,0.9169248,0.03182109,0.01567346],"genre_scores_gemma":[0.009612595,0.000286359,0.01081317,0.0005697733,0.0001589208,0.0003344346,0.9703819,0.002176113,0.005666687],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.9949433,"threshold_uncertainty_score":0.1041135,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02574690872258406,"score_gpt":0.261684851727758,"score_spread":0.235937943005174,"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."}}