{"id":"W4417003228","doi":"10.48550/arxiv.2512.02240","title":"Lightweight Latent Reasoning for Narrative Tasks","year":2025,"lang":"","type":"preprint","venue":"arXiv (Cornell University)","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Canadian Institute for Advanced Research","keywords":"Security token; Latent variable; Encoding (memory); Reinforcement learning; Matching (statistics); Debiasing; Language model; Reasoning system","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001789434,0.001433835,0.0008339361,0.0009979123,0.000666879,0.001731444,0.00318873,0.001457447,0.009201046],"category_scores_gemma":[0.008637699,0.000792008,0.001891645,0.0008400928,0.0007822821,0.004920064,0.002369558,0.003187623,0.004485847],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001273105,"about_ca_system_score_gemma":0.001427622,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00374028,"about_ca_topic_score_gemma":0.009176453,"domain_scores_codex":[0.9986038,0.0005897588,0.00008375997,0.0003790066,0.0002348524,0.0001088038],"domain_scores_gemma":[0.9965926,0.002116524,0.0002241362,0.0007325109,0.0001779864,0.0001561992],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0007746296,0.000529258,0.003788977,0.001038675,0.0002150987,0.0004132418,0.001017192,0.2273709,0.01860119,0.05033509,0.03049937,0.6654164],"study_design_scores_gemma":[0.00005785174,0.00005834747,0.0002979329,0.00003222374,0.00002778188,0.00007693755,0.00008515512,0.9355227,0.00483036,0.05346202,0.005526781,0.00002184617],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01257213,0.0005173605,0.9704315,0.0004099499,0.00005288223,0.0001236422,0.0007181583,0.01344291,0.001731423],"genre_scores_gemma":[0.361796,0.0005113704,0.623583,0.0003797103,0.0001158989,0.0003730726,0.005132477,0.001296657,0.006811886],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.009201046,"threshold_uncertainty_score":0.03078055,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08023349451410348,"score_gpt":0.2070643758157319,"score_spread":0.1268308813016285,"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."}}