{"paper":{"title":"AutoSFT: An Agentic System for Automated Data Research in Supervised Fine-Tuning","arxiv_id":"2610.22194","primary_cat":"cs.AI","submitted_at":"2026-10-15T16:44:34+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":0.96,"id":"s0","own":false,"role":"background","text":"Language model agents increasingly automate post-training, but when given the full training stack they converge on supervised fine-tuning and spend most effort revising training data.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.32,"main_claim_p":0.05,"quantitative":false,"quantitative_p":0.25,"role_probabilities":{"gap":0.02,"method":0.0,"result":0.02,"background":0.96,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":0.72,"id":"s1","own":false,"role":"gap","text":"Existing systems give the agent one kind of control, selecting rows from a fixed pool or calling a generator, whereas a human data engineer adjusts proportions, prunes, repairs, replaces and synthesizes the data.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.08,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.49,"role_probabilities":{"gap":0.72,"method":0.0,"result":0.0,"background":0.28,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":0.5,"id":"s2","own":false,"role":"contribution","text":"We design a data action space for these changes and ask whether an automated system can make the full range rather than one of them, and what structure that requires.","hedge":{"p":0.73,"label":"unclear"},"own_p":0.39,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.09,"role_probabilities":{"gap":0.47,"method":0.01,"result":0.0,"background":0.02,"limitation":0.0,"future_work":0.0,"contribution":0.5,"interpretation":0.0}},{"i":3,"p":0.87,"id":"s3","own":true,"role":"result","text":"We first run this loop with a single agent, and it reaches only one action, adjusting category proportions, while its context fills with execution output and is discarded mid-search.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"main_claim_p":0.03,"quantitative":true,"quantitative_p":0.57,"role_probabilities":{"gap":0.05,"method":0.02,"result":0.87,"background":0.0,"limitation":0.06,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":4,"p":0.66,"id":"s4","own":true,"role":"method","text":"We therefore introduce \\textbf{AutoSFT}, a multi-agent system where a meta agent separates execution from analysis, gathers proposals from independent angles, and calls a synthesis subagent for capabilities the pool cannot cover, with the model, the trainer, and the budget fixed within each comparison.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.92,"main_claim_p":0.08,"quantitative":false,"quantitative_p":0.12,"role_probabilities":{"gap":0.0,"method":0.66,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.34,"interpretation":0.0}},{"i":5,"p":1.0,"id":"s5","own":true,"role":"result","text":"Adding each component in turn helps: the executor halves context loss, the analysis subagents lift the best IFEval by $+6.5$, and synthesis adds $+3.4$ IFEval and $+2.9$ IFBench.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.97,"main_claim_p":0.08,"quantitative":true,"quantitative_p":0.99,"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":"The same loop transfers to code, where analysis adds $+3.7$ MBPP+ over a single agent that does not improve.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"main_claim_p":0.03,"quantitative":true,"quantitative_p":0.98,"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":0.94,"id":"s7","own":true,"role":"method","text":"Every result comes from runs that retrain from the base model each round, so each component's effect is individually attributable.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.09,"role_probabilities":{"gap":0.0,"method":0.94,"result":0.03,"background":0.0,"limitation":0.02,"future_work":0.0,"contribution":0.0,"interpretation":0.01}},{"i":8,"p":1.0,"id":"s8","own":true,"role":"interpretation","text":"Our findings suggest that automating SFT data research depends more on covering diverse data actions than on a stronger single action.","hedge":{"p":1.0,"label":"hedged_possibility"},"own_p":0.93,"main_claim_p":0.9,"quantitative":false,"quantitative_p":0.37,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":1.0}}]},"compiled":{"main_claim":{"p":1.0,"text":"Our findings suggest that automating SFT data research depends more on covering diverse data actions than on a stronger single action.","location":"abstract s8","hedge_strength":{"p":1.0,"label":"hedged possibility"}},"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":7091,"cost_usd_reported":0.0},"compiled_at":"2026-10-16","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.09,"states_limitation":0.53,"human_participants":0.05,"controlled_comparison":0.88,"randomized_assignment":0.02,"states_quantitative_result":0.99,"compares_with_prior_results":0.9,"mentions_code_or_data_availability":0.04},"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":0.98,"value":"method or tool","probabilities":{"review":0.0,"method_or_tool":0.98,"empirical_study":0.02,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":1.0,"text":"We therefore introduce \\textbf{AutoSFT}, a multi-agent system where a meta agent separates execution from analysis, gathers proposals from independent angles, and calls a synthesis subagent for capabilities the pool cannot cover, with the model, the trainer, and the budget fixed within each comparison.","location":"abstract s4"}},"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"}}}