One is the latency of starting up the "cold" lambda, one is the latency of actually starting it up and running your code. The latter is a moving target, subject

One is the latency of starting up the "cold" lambda, one is the latency of actually starting it up and running your code. The latter is a moving target, subject to the setup i.e. size of loaded code, available memory/CPU, etc, and subject to the actual computation and the latency of its dependencies e.g. network calls, database queries, etc.