Write a handoff doc for an AI-built project

Write a handoff document for a project built largely with AI assistance.

Sections:
1. What it does, and what it deliberately does not.
2. How to run it — setup, environment, external services.
3. Architecture as actually built, not as intended.
4. *Areas I do not fully understand* — named explicitly, with what I know about each and what I have not verified. This section is the point of the document.
5. Known fragility — what broke before, what is untested, what was patched rather than fixed.
6. Where to be careful, and what to verify before changing it.
7. Decisions made and why, where I know the why.

Rules:
- *Do not present uncertain areas as understood.* A handoff that overstates confidence is worse than no handoff — the recipient will trust the wrong parts.
- Distinguish "tested and works" from "appears to work".
- Do not document intent as if it were behavior. Where I only know what it was supposed to do, say that.
- Be concrete about what the next person should read first.