{"id":"W7077887124","doi":"10.48448/rdjn-9j55","title":"ACCORD: Closing the Commonsense Measurability Gap","year":2025,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"Geochemistry and Geologic Mapping","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Commonsense reasoning; Suite; Benchmark (surveying); Closing (real estate); Automated reasoning; Commonsense knowledge; Code (set theory)","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":[],"consensus_categories":[],"category_scores_codex":[0.0020207,0.0003019453,0.0002903948,0.0002260206,0.0006254604,0.0004005043,0.004220471,0.0001911225,0.0002332327],"category_scores_gemma":[0.001005449,0.0002056007,0.00008674631,0.001484586,0.001381125,0.0001655211,0.001348304,0.0005078896,0.00009304132],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009637223,"about_ca_system_score_gemma":0.001090628,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004257121,"about_ca_topic_score_gemma":0.0001723151,"domain_scores_codex":[0.9973148,0.0001322546,0.0003101871,0.0009451593,0.0007166312,0.0005810287],"domain_scores_gemma":[0.99698,0.0002864971,0.0002521599,0.002094985,0.0002691164,0.0001172037],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000007241163,0.0002441586,0.0008926575,0.0002362781,0.0000664845,0.00004887767,0.000497152,0.0003112796,0.001471312,0.2167615,0.6463469,0.1331161],"study_design_scores_gemma":[0.0001868032,0.00002567875,0.000274984,0.0002360493,0.00002070441,0.00003821,0.0001284579,0.02866885,0.001013148,0.03507683,0.9338083,0.0005219342],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.00002478134,0.0003007184,0.06522942,0.014615,0.001051557,0.0002977251,0.000008725298,0.0004589739,0.9180131],"genre_scores_gemma":[0.1442921,0.00006048311,0.03933754,0.002877246,0.0004813981,0.00003177299,0.00001204865,0.00004095013,0.8128664],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.2874614,"threshold_uncertainty_score":0.8384152,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03445118359345169,"score_gpt":0.2781933986903419,"score_spread":0.2437422150968902,"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."}}