QINLING v4 experiment (900 runs): 90 tasks × 5 language models × 2 hard-rule conditions

Description

This dataset is anonymized audit evidence for the QINLING v4 mechanism study, comprising 900 create-and-run observations from 90 frozen Chinese tasks tested across 5 models and 2 conditions (3 local and 2 hosted models), used to compare the effect of hard-rule activation on high-risk disposition boundaries. The data are organized at task and group levels, including observed terminal states, Governance state, Workflow W, strict E2E outcomes, integrity flags, paired-effect statistics, and hosted-model call audit metadata (with credentials removed). It is intended for manuscript review and methodological transparency, and does not represent field deployment results or human subject conclusions.

Categories

Funders

• Shaanxi Science and Technology Department

  Ministry of Science and Technology of the People's Republic of China

  Shaanxi, Xi'an

Licence