In the Napkin Math newsletter, I share articles reasoning from first principles about how systems work, and use these mental models to analyze the performance and characteristics of systems. I have found this skill to be invaluable when designing production systems.
Some examples:
How many writes per second can MySQL fundamentally do?
How fast can a neural network be trained?
How much might logging for an average website cost?
How much data can be transferred in the initial TCP congestion window?
Check out the archives for more.