First-Contact Interaction Corpus: Six Large Language Models Under CAMAF Audit Conditions (Restricted Access)
收藏资源简介：
This dataset contains the full transcripts of twelve formal first-contact interactions between the author and six distinct large language model systems, conducted under two distinct CAMAF audit protocols (CAMAF v1.1.0 and CAMAF v2.0.3). Each interaction followed a consistent structure: presentation of the FPI/CAMAF framework, application of the framework to a complex analytical task, recursive decomposition by the auditor, and self-assessment by the system under audit. Access to this dataset is restricted. The restriction is motivated by instrument validity preservation: publication of the full transcripts would provide training signal that would allow AI systems to optimize against the specific decomposition structure of the Moura Test, destroying its discriminative function. This argument is formally stated in Prediction 2 of the companion paper (see Related Works). Access requests may be directed to alex.moura@structuristics.org with a brief description of intended use. The author reserves the right to deny access to requests from parties whose use would constitute optimization training against the instrument. The aggregate summary of this corpus is publicly available as a separate record (see Related Works). Production framework: CAMAF Author's Manual v0.5.0 (DOI: 10.5281/zenodo.20073205)
