Knowledge Map
This is a working memory for a human and AI building things together. The public reading surface is deliberately flatter than the repository:
Foundations preserve what compounds. Systems explain how work gets done. Frontier notes are useful because they admit when they expire.
Start Here
| Layer | What belongs here | Entry |
|---|---|---|
| Foundations | Computer science, mathematics, and AI concepts that survive product cycles | Foundations |
| AI Systems | MCP, skills, harnesses, evaluation, and the interfaces around models | AI Systems |
| Frontier | Dated model/API snapshots and emerging capabilities | Frontier Radar |
| Practice | Reusable tools and personal workflows | Tools & Workflows |
| Domains | Quantitative finance, culture, and continuing interests | Quantitative Finance · Culture & Interests |
| Reference | Detailed legacy course notes kept searchable but off the main path | Reference Shelf |
How to Navigate
- Use this map when deciding what kind of knowledge you need.
- Open AI Systems or Frontier from the sidebar for direct access to every maintained note in those branches.
- Use search for an exact term, command, error, or old course fragment.
- Follow Connected Notes when one idea crosses layers or domains.
Publishing Rules
- A concept supplied by the owner is a discovery lead, not factual authority. Preserve the question, then verify terminology, current sources, alternatives, and limitations.
- One note should answer one durable question or support one recurring decision.
- Stable concepts do not absorb release news. Volatile facts record
verified_atandreview_after. - Publish an initial idea as
seedordeveloping; use the growth queue to add evidence, examples, contrasts, and connections before calling it mature. - Update a useful canonical note before creating a near-duplicate.
- Keep raw capture private until it becomes reusable knowledge.
- Archive or remove obsolete trivia; Git history is the recovery layer.
- Preserve valuable public routes even when the visible navigation changes.
tip
For current AI-assisted work, start with AI Systems. Go directly to bounded agent loops for workflow design or single-GPU local models for a hardware decision; use the dated Model & API Radar only when a current hosted-product decision matters.