Plasticity-stability 作为长期能力：巩固、遗忘边界、可逆可塑性
Evaluation modality
Spec-levelA spec-motivation / governance borrow. Evaluated by spec review + contract tests, not A/B or ablation.
- Primary owner
- —
- Phase-A verdict
- —
- Shadow profile
- —
- Source papers
- EWC 2016 + Wake-Sleep 2024 + SESLR 2025 + semi-parametric consolidation
- Specs
- docs/specs/continuum-memory.mddocs/specs/multi-timescale-learning.mddocs/specs/evidence_program.md
Blind spot (现状盲点)
CMA-3 处理 memory drift telemetry，但它更像异常报警；还缺一条正向能力：系统如何在不断写入、遗忘、巩固、回滚中保持可塑性与稳定性。长期 lifeform 不只是"不漂移"，还要能可逆地学习新关系、新偏好、新任务，同时保留早期重要能力。
Adoptable suggestions (可落地动作)
- 1.在 continuum-memory / multi-timescale-learning 中加入 plasticity-stability probes：A→B→A 偏好更新、旧承诺保留、新关系不污染旧关系。PROPOSED
Not a runnable A/B candidate — evaluated by the path above, not ablation.
- 2.将 EWC / wake-sleep / latent replay 作为 memory/controller 侧算法候选，不进入 live substrate fine-tune。PROPOSED
Not a runnable A/B candidate — evaluated by the path above, not ablation.
- 3.与 CMA-3 telemetry 结合：不仅报告 drift，也报告可塑性不足、过度稳定、过度遗忘。PROPOSED
Not a runnable A/B candidate — evaluated by the path above, not ablation.
Traceability
No plugins / runs linked yet. Scaffold a suggestion to start.
Expected benefit (预期收益)
- 把持续学习从"不要灾难性遗忘"升级为可验收 competence。 - 给 memory promotion / decay / rollback 的阈值设计提供 evidence。 - 支撑多 interlocutor 长期关系的身份隔离。
Cited paper (引用论文)
**EWC**（2016）、**Wake-Sleep Consolidated Learning**（2024）、**SESLR**（2025）、**semi-parametric memory consolidation**。详见 [`research/probe/02_axis_walkthrough.md`](../../research/probe/02_axis_walkthrough.md) A3 与 [`research/arxiv-survey-2026-05.md`](../../research/arxiv-survey-2026-05.md) §10。 ---