{"id":"W7131453612","doi":"10.1109/icamac67779.2025.11398733","title":"CodeSync: Multi-Agent System for Programming Assistance and Deterministic Thinking","year":2025,"lang":"","type":"article","venue":"","topic":"AI-based Problem Solving and Planning","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Algoma University","funders":"","keywords":"Python (programming language); Benchmark (surveying); Cognitive architecture; Facilitator; Software; Commonsense reasoning; Automated reasoning; Architecture; Model-based reasoning","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","sts","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.001700127,0.0004447039,0.0005388788,0.0002062692,0.001535711,0.001741128,0.0009066811,0.0002239463,0.000003213416],"category_scores_gemma":[0.0002012212,0.0004299117,0.00014704,0.0005065705,0.0001302267,0.0003350245,0.0004421852,0.000298971,0.000008211035],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002038596,"about_ca_system_score_gemma":0.0003889054,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006005706,"about_ca_topic_score_gemma":0.00004634788,"domain_scores_codex":[0.9966882,0.0001737519,0.0007886682,0.001150626,0.0003157564,0.0008830304],"domain_scores_gemma":[0.9975325,0.001044322,0.0003099986,0.0006792251,0.0002307301,0.0002032378],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0000930052,0.0001950603,0.008283711,0.007910924,0.0002492663,0.0001065435,0.004699755,0.0008871466,0.0001434665,0.1315826,0.0007133709,0.8451352],"study_design_scores_gemma":[0.001293188,0.0002327153,0.0006870691,0.004522434,0.0001484976,0.00003190121,0.0006180367,0.973233,0.0003771237,0.0003190564,0.01796985,0.0005671178],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.001572165,0.003842821,0.9894713,0.0006062172,0.00186137,0.00125328,0.00001636293,0.0004336441,0.0009428218],"genre_scores_gemma":[0.5768596,0.00001228126,0.4202097,0.0002378012,0.00004865567,0.0001147179,0.000003533399,0.0000154813,0.002498208],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9723459,"threshold_uncertainty_score":0.9998153,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02826895790284494,"score_gpt":0.2864204106948949,"score_spread":0.25815145279205,"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."}}