Editorial
CGFEV: Auditable LLM-Assisted Qualification of Configuration-Sensitive Manufacturing Evidence for Commercial Seating Design
@ARTICLE{10.4108/eetsis.14612, author={Xu Huang and Chuanzhi Chen}, title={CGFEV: Auditable LLM-Assisted Qualification of Configuration-Sensitive Manufacturing Evidence for Commercial Seating Design}, journal={EAI Endorsed Transactions on Scalable Information Systems}, volume={13}, number={6}, publisher={EAI}, journal_a={SIS}, year={2026}, month={10}, keywords={evidence qualification, product configuration, large language models, intelligent manufacturing, automated compliance checking, provenance}, doi={10.4108/eetsis.14612} }- Xu Huang
Chuanzhi Chen
Year: 2026
CGFEV: Auditable LLM-Assisted Qualification of Configuration-Sensitive Manufacturing Evidence for Commercial Seating Design
SIS
EAI
DOI: 10.4108/eetsis.14612
Abstract
Engineering rules often rely on dimensions copied from supplier documents. A value may look correct yet belong to another chair option, measurement object, or source that lacks a required condition. Contract-Guided Frozen-Evidence Verification (CGFEV) freezes assigned files and target configuration, while a Rule Card states what must be established. An LLM writes typed propositions and locators; deterministic Gate v0.2 controls release. We evaluated four frozen routes on commercial seating. Holdout-A contained 48 family-isolated natural cases, including 47 with binary Gold; Holdout-B contained 20 controlled pairs. On Holdout-A, B1 released 4/8 Gold-ELIGIBLE cases and falsely admitted 6/39 Gold-NOTELIGIBLE cases. R0, L1, and the text-only V1 prompt variant recorded no observed false admissions and released 0/8 eligible cases. R0 completed 47/47 binary cases, whereas L1 and V1 ended in ERROR on 12 and 14 cases, leaving substantive decisions for 35/47 and 33/47. Every route missed all six eligible members and all four required direction changes in Holdout-B. All 88 formal V1 records had empty visualassets fields, so these results provide no evidence about visual or multimodal qualification. Together, these results distinguish false admission, eligible release, and execution completion as separate evaluation dimensions. CGFEV provides an auditable admission boundary for configuration-sensitive evidence qualification.
Copyright © 2026 Xu Huang et al., licensed to EAI. This is an open access article distributed under the terms of the CC BY-NC-SA 4.0, which permits copying, redistributing, remixing, transformation, and building upon the material in any medium so long as the original work is properly cited.