What building taught me.
A working collection of insights extracted from the actual projects. Not motivational — operational. Each note points at the project evidence behind it.
Reliability beats cleverness
A profitable-looking opportunity is irrelevant if the underlying state is uncertain. Phoenix exists to make that sentence structurally true.
engineering02Economic correctness matters
A strategy can be technically correct and economically wrong. The job of an execution system is to be the second kind of correct.
systems03Production is part of the product
Deployment, rollback, observability, and reconciliation are product capabilities. Without them, "live" is not a meaningful word.
architecture04AI needs infrastructure
Models are not the product by themselves. Routing, context, evaluation, tools, and feedback loops are. Free Best Router and UEA are the same lesson at different layers.
ai05Fail-closed is different from fast
When execution has real consequences, uncertainty must reduce authority rather than trigger action.
systems06Product thinking changes engineering
Architecture should exist because of a user / business / system problem, not because a technology is fashionable.
product07One stable interface in front of a churning ecosystem
A good abstraction absorbs change so the caller does not have to. The OpenAI-compatible surface of Free Best Router is that abstraction for free LLMs.
architecture08Agents need a kernel, not a personality
Engineering agents accumulate tribal knowledge until they are unmaintainable. The fix is a small kernel plus a project-specific profile.
ai09Trust is a product
A trading system is not a strategy. It is the integration of market data, decision logic, risk gates, and a UI that lets an operator trust what the system is doing.
product10Small kernel, large surface
The right primitive is small enough to be obviously correct and large enough to be obviously useful. Phoenix, Free Router, UEA all share this shape.
architecture