Design principles (philosophy consistency)

Parent: product

Every design choice traces back to an articulated stance — that internal consistency is a hard requirement, not decoration.

Products are expression, not identity → the page doesn't sell who you are; it surfaces what you think.
Reject reference figures → no "I'm like X" framing anywhere.
Serve those who already love it (Strategy B) → depth-tilted content; keep some access friction (you have to chat to go deeper).
Network-shaped thinking → a chat interface replaces a linear narrative.
Reading-first philosophy → typography-driven; let substance speak.
5-year serial settler → don't brand yourself as "the founder of project X".
Architectural prediction — individual + AI subagent as the unit of ownership → StandMeet is an early instance of that future.

Principles the engineering produced

The list above was reasoned down from a stance. These four came the other way — each was learned by shipping something wrong, and each turned out to be a product decision rather than an implementation detail.

Never offer an action that cannot happen. A button with no handler, an upgrade the instance has no authority to perform, a tool named in a capability's instructions that this session was not granted. All of them fail silently — the visitor or owner acts and nothing occurs — and "is it visible" cannot tell the difference. Where the action genuinely cannot happen, the surface says so and says what to do instead (deployment).
A number that is always the same number is not information. The reader's meta row carried "N corpus sources" that read 0 on all 575 entries, because it counted a promotion chain the vault-import path never creates. The instance reported a version that was a literal in the source rather than the build it was running. Both looked like knowledge and cost the reader more than a blank would have.
State facts; let capabilities give instructions. The kernel tells the agent the time, the zone, which page the visitor is on. It does not tell it how to schedule — that arrives with the booking capability, so a visitor without it never reads scheduling advice. Product consequence: the AI never promises on behalf of a capability this visitor does not have.
The corpus answers; the model does not fill gaps. Measured against the real model: asked something the notes do not cover, it declines, names the nearest note it does have, and says it will not invent one. That behaviour is what makes "surfaces what you actually thought" a property rather than a hope — and it is the one place where a wrong answer is not a bug but a harm, since the answer carries the owner's name.