Building Reliable Full-Stack Systems

Reliability is usually discussed as an operations problem. In practice most of it is decided at design time, in choices that look like style preferences and are not.

Put the boundary where the trust changes

Authorisation belongs at the layer that owns the data, not at the layer that renders it. A UI that hides a button is a courtesy. An API that rejects the request is a boundary. Confusing the two is how a student project becomes an incident.

Prefer explicit state over inferred state

Deriving state is elegant until two derivations disagree. When something matters — an order status, a session, a job's progress — store it, name it, and make its transitions explicit.

Design for the second engineer

• ›Code is read far more often than it is written, and usually by someone with less context than you have right now.
• ›A boring, obvious implementation beats a clever one that needs a paragraph of explanation.
• ›If a decision was non-obvious, the reason belongs in the repository, not in your memory.