loading…
A prototype proves that one path can work. Production requires the system to keep producing an acceptable customer outcome across bad inputs, slow dependencies, deploys, restarts, concurrency, and operator handoffs.
Correctness describes one execution. Reliability describes behavior over time and under failure.
Ship → Observe → Detect → Recover → Improve
Use development to iterate, staging to rehearse production-like behavior, and production to serve real users under explicit ownership. “Who notices, who responds, and how do we recover?” are product questions.
Before launch, write down the owner, the first three failure signals, the safe degraded behavior, the recovery path, and the rollback path. If one is missing, the system is still operating like a demo.
Choose one customer workflow and draw every trust boundary, network call, state transition, and side effect. Then repeat the exercise in Kintsu Agent. Those marks become the incident surface for the rest of the course.