Latent Proxy-Evidence Auditing for Generative Data Mining
EvAudit is a generative evidence-auditing framework that filters unsupported semantic and macro-spatial claims before downstream ingestion. It combines proxy-supported evidence gates, calibration, stability checks, selective escalation, and auditable evaluation across generative and retrieval-style settings.
This repository contains the public research artifact: metrics, calibration state, configurations, validation utilities, example public pipelines, tests, figures, and reproducibility documentation. Data with external licensing or access constraints are referenced rather than silently redistributed.
Create an environment using either environment.yml or requirements.txt, then inspect verify_artifact.py, verify_reported_results.py, and the public validation scripts for the intended audit paths.
Chaewon Yoon
chaewon.yoon.ds@gmail.com
See LICENSE.txt and the per-asset documentation for applicable terms.