This page describes our own working card library — the pile of reading we keep in retrievable form. It is not a blog roll and not a republishing platform: it is the corpus we practise on before we touch a client's documents.
The library currently holds 200+ cards, drawn from Chinese technology, business and strategy publications. Every card holds one idea, its source, and a one-sentence takeaway written in our own words.
Why we keep a card library at all
Because retrieval quality is decided long before anyone types a question. Two failures cause most knowledge-base projects to disappoint:
- Documents are too coarse. A 40-page PDF answers twelve different questions. Retrieval hands the model the whole thing and hopes for the best.
- Provenance is missing. Nobody can say where a claim came from, so nobody can defend an answer when a customer challenges it.
A card library is the smallest unit that solves both: one idea per card, one source per card, one takeaway that can be checked against the original.
How the library is organised
| Shelf | What it holds |
|---|---|
| 🔧 Technology — AI tools and engineering practice | Agent frameworks, coding assistants, deployment notes, model evaluations, local-inference experience |
| 🏪 Business — operations, management, strategy | Company case studies, pricing and channel thinking, management frameworks, annual reviews |
| 🧠 Thinking — cognition and cross-domain models | Mental models, decision frameworks, reading lists, long-form interviews |
What a single card contains
| Field | Why it exists |
|---|---|
| Title | Written as the claim or the question — never a vague label |
| Source | The publication it came from, kept so the original can be found again |
| Shelf / category | So a search can be narrowed before it starts |
| One-sentence takeaway | The retrievable unit: a single idea, stated once, in plain language |
| Date | So older cards can be re-checked rather than silently trusted |
Why this is the same discipline we sell
When we build a client knowledge base, we apply exactly these habits at a larger scale:
- Chunk by idea, not by file. One answer per chunk, so retrieval returns one thing instead of a chapter.
- Keep metadata with the content. Version, owner, category, effective date — attached to the text, not kept in someone's head.
- Keep the source traceable. An answer that cannot be traced back to a document is an answer nobody will defend.
- Write for one question at a time. That is also how we write our answer library.
Cards, sources and rights
The cards are working notes on other people's published writing. We record the source, keep only a short takeaway of our own, and do not reproduce full texts. If you are a rights holder and want a card removed, contact us and we will remove it.
Reading it
- The full library, with all 200+ cards, is maintained in Chinese: 精选知识卡片库.
- For English readers, the equivalent in our own voice is the answer library — short, sourced answers on knowledge bases, AI customer service and AEO.
- For what the studio itself claims, see the fact sheet.
A card library is a cheap way to find out whether your organisation can keep knowledge retrievable at all. If nobody can maintain 200 cards, nobody will maintain 2,000 documents — better to learn that before buying a system.