{"id":"W4414991919","doi":"","title":"CrafterDojo: A Suite of Foundation Models for Building Open-Ended Embodied Agents in Crafter","year":2025,"lang":"en","type":"preprint","venue":"ArXiv.org","topic":"Human Motion and Animation","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kootenay Association for Science & Technology","funders":"Institute for Information and Communications Technology Promotion; Ministry of Science and ICT, South Korea; National Research Foundation of Korea; National Research Foundation","keywords":"Suite; Testbed; Embodied cognition; Key (lock); Foundation (evidence); Benchmark (surveying)","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0002613116,0.0002261257,0.0003672386,0.0002979534,0.00003829745,0.00008855573,0.0003814596,0.0002053795,0.00008636976],"category_scores_gemma":[0.00004129677,0.0002647427,0.0001101087,0.000118552,0.00001468169,0.0002940555,0.0003320504,0.0002476412,0.00001458829],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001886476,"about_ca_system_score_gemma":0.00004579537,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005021883,"about_ca_topic_score_gemma":0.000104871,"domain_scores_codex":[0.9987639,0.0000346684,0.0005386906,0.0003305177,0.0001197531,0.0002124275],"domain_scores_gemma":[0.999388,0.00004113552,0.0001109344,0.0003293414,0.00009024042,0.00004031749],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001862285,0.0003362882,0.04569122,0.006501217,0.0004185368,0.000007155464,0.004639405,0.9019654,0.01684879,0.01111504,0.003336351,0.008954395],"study_design_scores_gemma":[0.00186687,0.00003215149,0.0916109,0.001598209,0.00007034069,4.297251e-7,0.00005099938,0.8793336,0.003305686,0.01957397,0.001978178,0.0005786206],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8696724,0.00004381996,0.1227481,0.00008284993,0.0005583194,0.001130148,0.00005067865,0.0001278788,0.00558583],"genre_scores_gemma":[0.9953954,0.00004154268,0.003115825,0.00008654372,0.00006489008,0.000319087,0.0001497342,0.00003203366,0.0007949323],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.125723,"threshold_uncertainty_score":0.9999804,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1086453667221213,"score_gpt":0.3360080037629529,"score_spread":0.2273626370408316,"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."}}