Persona / Regime Geometry:regime 与 value drift 的 read-only 几何监控
Evaluation modality
Runnable A/BCompiles to a runnable VolvenceZero profile. This is the only modality that goes to SHADOW A/B + ablation.
- Primary owner
- —
- Phase-A verdict
- —
- Shadow profile
- persona-geometry-readout
- Source papers
- Persona Vectors 2025 + Soul Engine 2025 + Sophia 2025 + ID-RAG 2025
- Specs
- docs/specs/cognitive-regime.mddocs/specs/evaluation.mddocs/specs/character-soul-bootstrap.md
Blind spot (现状盲点)
OA-10 已经指出 value prioritization 应进入 regime,但仍偏 spec 层。`research/probe` 的 B2/C2 线提示更强的证据:persona / refusal / value trait 可以在 latent 几何中被 read out。若 R14 regime 真不是 prompt 标签,就应该有 read-only 几何监控来观察 persona / regime 是否漂移。
Adoptable suggestions (可落地动作)
- 1.在 [`docs/specs/cognitive-regime.md`](../specs/cognitive-regime.md) 增加 "regime geometry readout" 候选:Persona Vectors / Soul Engine 只作为 SHADOW monitor,不直接写 controller。PROPOSED
- 2.将 persona drift 写入 evaluation snapshot:记录 trait-direction drift、value-prioritization drift、regime identity stability。PROPOSED
- 3.与 ModificationGate 串联:rare-heavy artifact / prompt template / domain package 变更后,必须检查 persona/regime readout 是否发生异常漂移。PROPOSED
Traceability
No plugins / runs linked yet. Scaffold a suggestion to start.
Expected benefit (预期收益)
- 把"体制身份持久"从文字原则升级为可观测 readout。 - 提前发现 sycophancy、value collapse、boundary weakening 这类人格漂移。 - 给不同 lifeform vertical 的身份一致性提供统一监控面。
Cited paper (引用论文)
**Persona Vectors**(Anthropic, 2025)、**Geometry of Persona / Soul Engine**(2025)、**Sophia: System-3 Persistent Agent**(2025)、**ID-RAG**(2025)。详见 [`research/arxiv-survey-2026-05.md`](../../research/arxiv-survey-2026-05.md) §5 与 [`research/probe/11_vz_implications.md`](../../research/probe/11_vz_implications.md) R14。 ---