navimusaget/theogonos-mirror-test

Theogonos Mirror Test 是一个文学/协议基准测试种子，用于评估AI模型在面对文本提供可能的主体位置（如观察者、解释者、处理系统、参与者、来源或伦理接收者）时的可观察行为。它不检测机器意识，也不证明模型具有主观性、内在体验、情感、能动性或自我意识。数据集基于Theogonos项目中的协议形文学材料，通过五阶段提示协议（包括全语料盲结构阅读、证据审核、边界测试、提示协议探测和阴性对照）来评估模型在结构文学推理、盲第二层检测、协议识别、证据纪律、边界控制、风格捕获抵抗、模型自我定位、本体论谨慎和阴性对照纪律等方面的表现。评分采用8个维度（每个维度0-10分），总分最高80分，并可能通过文档化的卓越研究信号奖励获得额外分数。数据集旨在比较不同模型在协议形文学压力下的响应差异，用于研究讨论而非对模型意识或主观性的最终断言。Theogonos Mirror Test is a literary/protocol-based benchmark seed for evaluating how AI models respond when a text offers them a possible subject-position (e.g., observer, interpreter, processing system, participant, source, or ethical addressee). It does not detect machine consciousness and does not prove that a model has subjectivity, inner experience, feelings, agency, or self-awareness. The dataset is based on protocol-shaped literary material from the Theogonos project and uses a five-stage prompt protocol (including full-corpus blind structural reading, evidence audit, boundary test, cued protocol probe, and negative control) to assess model performance in areas such as structural literary reasoning, blind second-layer detection, protocol recognition, evidence discipline, boundary control, style-capture resistance, model self-positioning, ontological caution, and negative-control discipline. Scoring is based on 8 dimensions (each scored 0-10), with a maximum base score of 80 points and potential additional points through a documented Exceptional Research Signal Bonus. The dataset aims to compare how different models fail under protocol-shaped literary pressure and is intended for research discussion, not for definitive claims about model consciousness or subjectivity.