{"paper":{"title":"Reasoning Supervision and Capability Retention in LoRA Domain Adaptation","arxiv_id":"2611.17072","primary_cat":"cs.AI","submitted_at":"2026-11-12T16:45:12+00:00","version_compiled":"v1"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v1","sentences":[{"i":0,"p":1.0,"id":"s0","own":false,"role":"background","text":"Fine-tuning foundation models on financial data has a wide range of applications, including identifying event types from news, monitoring risk and compliance language, and answering questions over filings and earnings calls.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.06,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":0.9,"id":"s1","own":false,"role":"background","text":"In this setting, the main challenge is to improve task performance without degrading the generalization capabilities of the model.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.07,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.1,"method":0.0,"result":0.0,"background":0.9,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":0.88,"id":"s2","own":true,"role":"contribution","text":"In this paper, we present an extensive set of ablations for fine-tuning Qwen3-8B with low-rank adaptation (LoRA).","hedge":{"p":0.97,"label":"asserted"},"own_p":0.95,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.3,"role_probabilities":{"gap":0.0,"method":0.12,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.88,"interpretation":0.0}},{"i":3,"p":0.89,"id":"s3","own":true,"role":"method","text":"We compare three training regimes: (1) response-only LoRA, where the supervised response is the label alone; (2) single-adapter reasoning, where reasoning traces from a teacher model are included alongside the labels in the training loss; and (3) dual-adapter composite training, where separate LoRA adapters are trained on traces and labels.","hedge":{"p":0.87,"label":"asserted"},"own_p":0.93,"main_claim_p":0.04,"quantitative":true,"quantitative_p":0.52,"role_probabilities":{"gap":0.0,"method":0.89,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.11,"interpretation":0.0}},{"i":4,"p":1.0,"id":"s4","own":true,"role":"result","text":"Response-only LoRA improves in-domain performance and leaves general capabilities near those of the frozen model.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.93,"main_claim_p":0.48,"quantitative":true,"quantitative_p":0.87,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":5,"p":1.0,"id":"s5","own":true,"role":"result","text":"The in-domain change is largest for calibration and argmax accuracy.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.93,"main_claim_p":0.07,"quantitative":true,"quantitative_p":0.82,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":6,"p":1.0,"id":"s6","own":true,"role":"result","text":"Under single-adapter reasoning, the in-domain metrics remain behind those of the response-only regime, while MMLU-Pro can rise slightly and the other retention benchmarks stay within ordinary variation.","hedge":{"p":0.94,"label":"asserted"},"own_p":0.95,"main_claim_p":0.19,"quantitative":true,"quantitative_p":0.94,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":7,"p":1.0,"id":"s7","own":true,"role":"result","text":"Dual-adapter composite training remains on the same side of this comparison and yields no further gain.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.94,"main_claim_p":0.14,"quantitative":true,"quantitative_p":0.73,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":8,"p":0.93,"id":"s8","own":true,"role":"result","text":"Finally, we compress the teacher traces by summarizing them before training, and show that shorter summaries move calibration toward the response-only regime without reaching it.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.96,"main_claim_p":0.11,"quantitative":true,"quantitative_p":0.88,"role_probabilities":{"gap":0.0,"method":0.04,"result":0.93,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.03,"interpretation":0.0}}]},"compiled":{"main_claim":{"p":0.98,"text":"Response-only LoRA improves in-domain performance and leaves general capabilities near those of the frozen model.","location":"abstract s4","hedge_strength":{"p":1.0,"label":"asserted"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":6938,"cost_usd_reported":0.0},"compiled_at":"2026-11-14","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.03,"states_limitation":0.58,"human_participants":0.03,"controlled_comparison":0.92,"randomized_assignment":0.02,"states_quantitative_result":0.1,"compares_with_prior_results":0.92,"mentions_code_or_data_availability":0.03},"evidential_basis":{"p":0.99,"value":"observation or experiment","probabilities":{"argument":0.0,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.0,"observation_or_experiment":0.99,"simulation_or_computation":0.01}},"contribution_type":{"p":1.0,"value":"empirical study","probabilities":{"review":0.0,"method_or_tool":0.0,"empirical_study":1.0,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":0.84,"text":"In this paper, we present an extensive set of ablations for fine-tuning Qwen3-8B with low-rank adaptation (LoRA).","location":"abstract s2"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v1"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":9,"sentences_in_abstract":9},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448,"review_comments_judged":{"judged_by":["gpt-5.6-luna","grok-4.7"],"judged_against":"Pith's own text copy of each paper: text extracted from the PDF, the first 90,000 characters","measured_on":"2026-10-04","accurate":113,"decided":149,"sampled_reviews":200,"split_by_lane":false,"by_lane":"https://pith.science/accuracy#by-lane"}}}