CIFAR Synthetic Evidence Corpus

A corpus and benchmark for detecting AI-manipulated evidentiary documents, organized by manipulation sophistication and generator family, with controls that let evaluators distinguish genuine forgery signals from shortcut learning.本语料库与基准测试集旨在检测AI篡改的证据文档，依据篡改复杂度与生成器家族进行分类组织，并配备对照机制，可使评估者区分真实的伪造信号与捷径学习所产生的干扰信号。