Mental models: building the latticework

“You’ve got to have models in your head… and you’ve got to array your experience — both vicarious and direct — on this latticework of models.” That’s Charlie Munger in A Lesson on Elementary, Worldly Wisdom (USC, 1994), the talk that started the whole genre. Read it before any book on the subject — it is the primary source and still the best.

The canon

• The Great Mental Models, volumes 1–4 (Farnam Street / Portfolio) — the most systematic treatment: general thinking concepts, then models from physics, chemistry, biology, systems, math, economics. One model per chapter, each with failure modes — which is what separates it from listicles. Volume 1 is the right entry point.
• Poor Charlie’s Almanack — the latticework demonstrated rather than described, across decades of talks.
• Seeking Wisdom by Peter Bevelin — drier than the others but the most complete inventory of misjudgment, drawn from Darwin, Munger, and Feynman.
• The FS mental models index — the free reference. Bookmark it; use it as a checklist when a decision feels stuck, not as reading material.

The models that earn their keep in engineering

A model you can’t apply is trivia. These are the ones I reach for weekly:

• Inversion — debug by asking what would have to be true for this failure to occur.
• Second-order effects — every cache, retry, and fallback you add has behavior under failure that is the real design question.
• Margin of safety — from engineering via Graham: the interesting question is never the expected load, it’s the load at which things break and what happens then.
• Base rates — “how often does this kind of project ship on time” beats any amount of inside-view planning detail.
• Map is not the territory — dashboards, benchmarks, and evals are maps. Every project I’ve built has found at least one place where the map lied.
• Circle of competence — the honest edge of what you know. The practical version: write down what you’d bet on vs. what you’d merely argue for.

How to actually internalize them

Reading produces recognition, not fluency. What worked for me: after any significant decision or postmortem, name the model that applied (or would have helped) in one line. A model you have used five times on real problems is worth fifty you can define.