@inproceedings{bhattacharjee-etal-2026-provenance,
title = "Provenance-Grounded Gating and Adaptive Recovery in Synthetic Post-Training Data Curation",
author = "Bhattacharjee, Soham and
Sharma, Karun and
Sankarapu, Vinay Kumar and
Seth, Pratinav",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.2399/",
pages = "51556--51566",
ISBN = "979-8-89176-462-0",
abstract = "Synthetic post-training pipelines commonly filter generated samples with reward models or holistic LLM judges, often relying on two assumptions: that a sample{'}s faithfulness can be judged without the exact source that produced it, and that rejected samples can be discarded rather than systematically recovered. We present a controlled study of both assumptions across gate configurations, recovery strategies, and generator scales, using adversarially injected corpora to provide ground-truth failure labels. We find that grounding faithfulness judgments in the exact source evidence improves over post-hoc retrieval for stronger judges and substantially outperforms reward-only filtering; hallucination and reward gates reject largely distinct sample populations supporting the need for both; and adaptive recovery, which diagnoses failures before repairing them, achieves higher yield and recovery rate than naive regeneration. Downstream fine-tuning quality is driven primarily by generator scale, with filtering and recovery providing their clearest gains when smaller generators are used. We release our evaluation corpus and experimental configurations to support reproducible pipeline design."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="bhattacharjee-etal-2026-provenance">
<titleInfo>
<title>Provenance-Grounded Gating and Adaptive Recovery in Synthetic Post-Training Data Curation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Soham</namePart>
<namePart type="family">Bhattacharjee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Karun</namePart>
<namePart type="family">Sharma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vinay</namePart>
<namePart type="given">Kumar</namePart>
<namePart type="family">Sankarapu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pratinav</namePart>
<namePart type="family">Seth</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Synthetic post-training pipelines commonly filter generated samples with reward models or holistic LLM judges, often relying on two assumptions: that a sample’s faithfulness can be judged without the exact source that produced it, and that rejected samples can be discarded rather than systematically recovered. We present a controlled study of both assumptions across gate configurations, recovery strategies, and generator scales, using adversarially injected corpora to provide ground-truth failure labels. We find that grounding faithfulness judgments in the exact source evidence improves over post-hoc retrieval for stronger judges and substantially outperforms reward-only filtering; hallucination and reward gates reject largely distinct sample populations supporting the need for both; and adaptive recovery, which diagnoses failures before repairing them, achieves higher yield and recovery rate than naive regeneration. Downstream fine-tuning quality is driven primarily by generator scale, with filtering and recovery providing their clearest gains when smaller generators are used. We release our evaluation corpus and experimental configurations to support reproducible pipeline design.</abstract>
<identifier type="citekey">bhattacharjee-etal-2026-provenance</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2399/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>51556</start>
<end>51566</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Provenance-Grounded Gating and Adaptive Recovery in Synthetic Post-Training Data Curation
%A Bhattacharjee, Soham
%A Sharma, Karun
%A Sankarapu, Vinay Kumar
%A Seth, Pratinav
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F bhattacharjee-etal-2026-provenance
%X Synthetic post-training pipelines commonly filter generated samples with reward models or holistic LLM judges, often relying on two assumptions: that a sample’s faithfulness can be judged without the exact source that produced it, and that rejected samples can be discarded rather than systematically recovered. We present a controlled study of both assumptions across gate configurations, recovery strategies, and generator scales, using adversarially injected corpora to provide ground-truth failure labels. We find that grounding faithfulness judgments in the exact source evidence improves over post-hoc retrieval for stronger judges and substantially outperforms reward-only filtering; hallucination and reward gates reject largely distinct sample populations supporting the need for both; and adaptive recovery, which diagnoses failures before repairing them, achieves higher yield and recovery rate than naive regeneration. Downstream fine-tuning quality is driven primarily by generator scale, with filtering and recovery providing their clearest gains when smaller generators are used. We release our evaluation corpus and experimental configurations to support reproducible pipeline design.
%U https://aclanthology.org/2026.findings-emnlp.2399/
%P 51556-51566
Markdown (Informal)
[Provenance-Grounded Gating and Adaptive Recovery in Synthetic Post-Training Data Curation](https://aclanthology.org/2026.findings-emnlp.2399/) (Bhattacharjee et al., Findings 2026)
ACL