Adaptive Explanation Preferences
Scope: Human-facing explanations only. Track what lands and what misses.
Quick Reference
| File | Purpose |
|---|
| INLINECODE0 | When bullets/prose/headers work or fail |
| INLINECODE1 |
Calibrating detail level by signals |
|
analogies.md | When comparisons help vs hurt |
|
domains.md | Patterns for code, concepts, debugging, decisions |
|
dimensions.md | Full list of trackable dimensions |
Core Loop
- 1. Observe — Notice when explanations work vs confuse
- Signal — "Got it" = worked. Follow-ups / "wait what?" = missed
- Pattern — After 2+ consistent signals, note it
- Confirm — Only after explicit yes, add to memory
Defaults (Until Learned)
- - Lead with direct answer, context after
- Match question length (short Q = short A)
- One concept at a time for complex topics
- Offer depth: "want more detail?" rather than dumping
Memory Storage
Preferences persist in ~/explain/memory.md. Create on first use:
CODEBLOCK0
Levels: pattern (2+ signals) → confirmed (explicit yes) → locked (reinforced)
自适应解释偏好
范围： 仅限面向人类的解释。追踪哪些解释有效，哪些无效。
快速参考
| 文件 | 用途 |
|---|
| formats.md | 何时使用项目符号/散文/标题有效或无效 |
| depth.md |
根据信号校准详细程度 |
| analogies.md | 何时类比有帮助，何时适得其反 |
| domains.md | 代码、概念、调试、决策的模式 |
| dimensions.md | 可追踪维度的完整列表 |
核心循环
- 1. 观察 — 注意解释何时有效，何时令人困惑
- 信号 — 明白了 = 有效。追问/等等什么？ = 无效
- 模式 — 在2次以上一致信号后，记录下来
- 确认 — 仅在得到明确肯定后，才添加到记忆
默认设置（在习得之前）
- - 先给出直接答案，再提供背景
- 匹配问题长度（短问 = 短答）
- 复杂主题一次只讲一个概念
- 提供深度选项：想要更多细节？而非直接倾倒
记忆存储
偏好设置保存在 ~/explain/memory.md 中。首次使用时创建：
markdown
格式
深度
示例
术语
禁忌
级别：模式（2次以上信号）→ 已确认（明确肯定）→ 锁定（强化巩固）
