R8 proposes a structural approach to approximating cognitive manipulation risk in Japanese-language text across 12 theoretically grounded categories. This is an exploratory framework at an early stage of empirical development; the claims made in this paper are limited to what the current implementation can support. The system produces a single composite score, the Cognitive Manipulation Index (CMI). Calibration against a 208-document corpus yields Precision=97.7%, Recall=35.5%, F1=52.1% (standard mode, threshold≥41). Inter-rater agreement between the automated system and an LLM pilot yielded Cohen's κ=0.211 (Fair, n=204), interpreted as evidence that the annotation protocol's structural reading load exceeds current LLM capability rather than a failure of the lexical detection system. The theoretical framework integrates cognitive psychology (Kahneman, 2011), social psychology (Janis, 1972; Milgram, 1963), and Japanese organizational sociology (Yamamoto, 1977; Nakane, 1967). The complete corpus and scoring data are publicly available at https://github.com/takahiro-oss/r8-cognitive-risk under CC BY 4.0.
Takahiro Saito (Sun,) studied this question.