{"id":"https://openalex.org/W7204238391","doi":"https://doi.org/10.48550/arxiv.2608.23911","title":"PROOF-Gen: From Optimized Data to Better Distillation","display_name":"PROOF-Gen: From Optimized Data to Better Distillation","publication_year":2026,"publication_date":"2026-08-24","ids":{"openalex":"https://openalex.org/W7204238391","doi":"https://doi.org/10.48550/arxiv.2608.23911"},"language":null,"primary_location":{"id":"pmh:oai:arXiv.org:2608.23911","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.23911","pdf_url":"https://arxiv.org/pdf/2608.23911","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2608.23911","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5134977949","display_name":"Anh Ta","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ta, Anh","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100698979","display_name":"Junjie Zhu","orcid":"https://orcid.org/0000-0001-7177-8694","observed_orcids":["https://orcid.org/0000-0001-7177-8694"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhu, Junjie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5061162179","display_name":"Shahin Shayandeh","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shayandeh, Shahin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.385699987411499,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.385699987411499,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.11110000312328339,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.1062999963760376,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/execution-trace-analysis","display_name":"execution trace analysis","score":0.5230000019073486},{"id":"https://openalex.org/keywords/on-device-deployment","display_name":"on-device deployment","score":0.6209999918937683},{"id":"https://openalex.org/keywords/llm-fine-tuning","display_name":"LLM fine-tuning","score":0.6240000128746033}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7013000249862671},{"id":"https://openalex.org/C13662910","wikidata":"https://www.wikidata.org/wiki/Q193139","display_name":"Trajectory","level":2,"score":0.641700029373169},{"id":"https://openalex.org/C175309249","wikidata":"https://www.wikidata.org/wiki/Q725864","display_name":"Pipeline transport","level":2,"score":0.4952000081539154},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.4821999967098236},{"id":"https://openalex.org/C173246807","wikidata":"https://www.wikidata.org/wiki/Q7833062","display_name":"Trajectory optimization","level":3,"score":0.4788999855518341},{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.4537999927997589},{"id":"https://openalex.org/C75291252","wikidata":"https://www.wikidata.org/wiki/Q1315756","display_name":"TRACE (psycholinguistics)","level":2,"score":0.45100000500679016},{"id":"https://openalex.org/C2776650193","wikidata":"https://www.wikidata.org/wiki/Q264661","display_name":"Obstacle","level":2,"score":0.44600000977516174},{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.43540000915527344},{"id":"https://openalex.org/C2776175482","wikidata":"https://www.wikidata.org/wiki/Q1195816","display_name":"Transfer (computing)","level":2,"score":0.41749998927116394},{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.37130001187324524},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3709999918937683},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.3707999885082245},{"id":"https://openalex.org/C2780337265","wikidata":"https://www.wikidata.org/wiki/Q185958","display_name":"Float (project management)","level":2,"score":0.33149999380111694},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.3314000070095062},{"id":"https://openalex.org/C182306322","wikidata":"https://www.wikidata.org/wiki/Q1779371","display_name":"Order (exchange)","level":2,"score":0.32179999351501465},{"id":"https://openalex.org/C61270487","wikidata":"https://www.wikidata.org/wiki/Q476843","display_name":"Interception","level":2,"score":0.319599986076355},{"id":"https://openalex.org/C48103436","wikidata":"https://www.wikidata.org/wiki/Q599031","display_name":"State (computer science)","level":2,"score":0.3174999952316284},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.314300000667572},{"id":"https://openalex.org/C2779538338","wikidata":"https://www.wikidata.org/wiki/Q2990590","display_name":"Completion (oil and gas wells)","level":2,"score":0.3111000061035156},{"id":"https://openalex.org/C44154836","wikidata":"https://www.wikidata.org/wiki/Q45045","display_name":"Simulation","level":1,"score":0.30329999327659607},{"id":"https://openalex.org/C29302406","wikidata":"https://www.wikidata.org/wiki/Q154411","display_name":"Wrench","level":2,"score":0.29490000009536743},{"id":"https://openalex.org/C2778415886","wikidata":"https://www.wikidata.org/wiki/Q1823435","display_name":"Reflector (photography)","level":3,"score":0.289000004529953},{"id":"https://openalex.org/C8505890","wikidata":"https://www.wikidata.org/wiki/Q605095","display_name":"Budget constraint","level":2,"score":0.28850001096725464},{"id":"https://openalex.org/C2780598303","wikidata":"https://www.wikidata.org/wiki/Q65921492","display_name":"Flexibility (engineering)","level":2,"score":0.28619998693466187},{"id":"https://openalex.org/C200601418","wikidata":"https://www.wikidata.org/wiki/Q2193887","display_name":"Reliability engineering","level":1,"score":0.2854999899864197},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.2816999852657318},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.27880001068115234},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.260699987411499}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.23911","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.23911","pdf_url":"https://arxiv.org/pdf/2608.23911","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2608.23911","doi":"https://doi.org/10.48550/arxiv.2608.23911","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.23911","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2608.23911","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.23911","pdf_url":"https://arxiv.org/pdf/2608.23911","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[{"display_name":"Quality education","score":0.4156000018119812,"id":"https://metadata.un.org/sdg/4"}],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7204238391.pdf","grobid_xml":"https://content.openalex.org/works/W7204238391.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Supervised":[0],"fine-tuning":[1],"on":[2,27,158,174],"teacher-generated":[3],"trajectories":[4,98],"is":[5,42,133],"the":[6,34,40,45,50,57,113,125,138,159,182],"standard":[7],"first":[8],"stage":[9,26],"for":[10],"distilling":[11],"tool-calling":[12,22],"capabilities":[13],"into":[14],"deployable":[15],"models.":[16],"Post-training":[17],"pipelines":[18],"that":[19,123],"drive":[20],"shipped":[21],"agents":[23],"re-run":[24],"this":[25],"a":[28,110,128,179,194],"daily":[29],"or":[30],"weekly":[31],"cadence,":[32],"paying":[33],"frontier-teacher":[35],"cost":[36],"each":[37,53,107],"cycle,":[38],"yet":[39],"mechanism":[41],"generate-and-filter":[43],"(keep":[44],"teacher's":[46],"passing":[47,129],"trajectories,":[48],"discard":[49],"rest)":[51],"and":[52,116,168,191],"cycle":[54],"leaves":[55],"behind":[56],"same":[58],"hard":[59],"scenarios":[60],"because":[61],"failures":[62,101],"supply":[63],"no":[64,145],"signal.":[65],"On":[66,148],"\u03c42-bench,":[67,149],"57%":[68],"of":[69,74,154],"teacher":[70,126],"trials":[71],"fail,":[72],"two-thirds":[73],"them":[75],"near-misses":[76],"(most":[77],"tool":[78],"calls":[79],"correct,":[80],"undone":[81],"by":[82,187],"one":[83],"decisive":[84],"error).":[85],"We":[86],"introduce":[87],"PROOF-Gen":[88],"(Per-scenario":[89],"Reflective":[90],"Optimization":[91],"to":[92,127,166,193,202],"Overcome":[93],"FailedGeneration),":[94],"which":[95],"recovers":[96,152],"golden":[97],"from":[99,141,164],"these":[100],"via":[102],"per-scenario":[103,150],"prompt":[104],"optimization.":[105],"For":[106],"failed":[108,155],"task,":[109],"reflector":[111],"analyzes":[112],"execution":[114],"trace":[115],"evaluation":[117],"feedback,":[118],"then":[119],"writes":[120],"corrective":[121],"guidance":[122,132],"steers":[124],"trajectory.":[130],"The":[131],"stripped":[134],"before":[135],"training,":[136],"so":[137],"student":[139],"learns":[140],"clean":[142],"demonstrations":[143],"with":[144,207],"task-specific":[146],"scaffold.":[147],"optimization":[151],"93%":[153],"scenarios.":[156],"Fine-tuned":[157],"combined":[160],"data,":[161],"Qwen3-4B-Instruct-2507":[162],"improves":[163],"Pass^1=0.132":[165],"0.529":[167],"Gemma":[169],"4":[170],"E4B-it":[171],"gains":[172],"+7.2pp":[173],"BFCL":[175],"v4":[176],"multi-turn.":[177],"In":[178],"deployed":[180,195],"pipeline,":[181],"method":[183],"lifts":[184],"trajectory":[185],"quality":[186],"+6.3pp":[188],"goal":[189,199],"completion":[190],"transfers":[192],"on-device":[196],"model":[197],"(+1.5pp":[198],"completion;":[200],"+1.7":[201],"+5.0pp":[203],"across":[204],"response-quality":[205],"metrics),":[206],"positive":[208],"transfer":[209],"in":[210],"every":[211],"locale":[212],"(non-English":[213],"average":[214],"+1.48pp).":[215]},"counts_by_year":[],"updated_date":"2026-10-14T16:39:01.820549","created_date":"2026-08-27T00:00:00"}