{"paper":{"title":"PROOF-Gen: From Optimized Data to Better Distillation","arxiv_id":"2608.23911","primary_cat":"cs.AI","submitted_at":"2026-08-24T23:33:56+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":"Supervised fine-tuning on teacher-generated trajectories is the standard first stage for distilling tool-calling capabilities into deployable models.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.05,"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":0.5,"id":"s1","own":false,"role":"background","text":"Post-training pipelines that drive shipped tool-calling agents re-run this stage on a daily or weekly cadence, paying the frontier-teacher cost each cycle, yet the mechanism is generate-and-filter (keep the teacher's passing trajectories, discard the rest) and each cycle leaves behind the same hard scenarios because failures supply no signal.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.17,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.18,"role_probabilities":{"gap":0.5,"method":0.0,"result":0.0,"background":0.5,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":0.71,"id":"s2","own":true,"role":"result","text":"On {\\tau}2-bench, 57% of teacher trials fail, two-thirds of them near-misses (most tool calls correct, undone by one decisive error).","hedge":{"p":1.0,"label":"asserted"},"own_p":0.81,"main_claim_p":0.03,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.08,"method":0.0,"result":0.71,"background":0.21,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":3,"p":0.97,"id":"s3","own":true,"role":"contribution","text":"We introduce PROOF-Gen (Per-scenario Reflective Optimization to Overcome FailedGeneration), which recovers golden trajectories from these failures via per-scenario prompt optimization.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.91,"main_claim_p":0.19,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.03,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.97,"interpretation":0.0}},{"i":4,"p":1.0,"id":"s4","own":true,"role":"method","text":"For each failed task, a reflector analyzes the execution trace and evaluation feedback, then writes corrective guidance that steers the teacher to a passing trajectory.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.88,"main_claim_p":0.05,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":1.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":5,"p":0.99,"id":"s5","own":true,"role":"method","text":"The guidance is stripped before training, so the student learns from clean demonstrations with no task-specific scaffold.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.83,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.99,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.01,"interpretation":0.0}},{"i":6,"p":1.0,"id":"s6","own":true,"role":"result","text":"On {\\tau}2-bench, per-scenario optimization recovers 93% of failed scenarios.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"main_claim_p":0.17,"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":7,"p":1.0,"id":"s7","own":true,"role":"result","text":"Fine-tuned on the combined data, Qwen3-4B-Instruct-2507 improves from Pass^1=0.132 to 0.529 and Gemma 4 E4B-it gains +7.2pp on BFCL v4 multi-turn.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.94,"main_claim_p":0.26,"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":8,"p":1.0,"id":"s8","own":true,"role":"result","text":"In a deployed pipeline, the method lifts trajectory quality by +6.3pp goal completion and transfers to a deployed on-device model (+1.5pp goal completion; +1.7 to +5.0pp across response-quality metrics), with positive transfer in every locale (non-English average +1.48pp).","hedge":{"p":1.0,"label":"asserted"},"own_p":0.92,"main_claim_p":0.39,"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}}]},"compiled":{"main_claim":{"p":0.84,"text":"In a deployed pipeline, the method lifts trajectory quality by +6.3pp goal completion and transfers to a deployed on-device model (+1.5pp goal completion; +1.7 to +5.0pp across response-quality metrics), with positive transfer in every locale (non-English average +1.48pp).","location":"abstract s8","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":7215,"cost_usd_reported":0.0},"compiled_at":"2026-09-23","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.09,"states_limitation":0.14,"human_participants":0.03,"controlled_comparison":0.83,"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":0.98,"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":0.98,"simulation_or_computation":0.02}},"contribution_type":{"p":1.0,"value":"method or tool","probabilities":{"review":0.0,"method_or_tool":1.0,"empirical_study":0.0,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":0.99,"text":"We introduce PROOF-Gen (Per-scenario Reflective Optimization to Overcome FailedGeneration), which recovers golden trajectories from these failures via per-scenario prompt optimization.","location":"abstract s3"}},"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"}}}