---
title: "How to Think From First Principles: Find Root Nodes"
description: "How to think from first principles: trace any inherited conclusion down to its root nodes, keep only what survives \"how do we know?\", and rebuild from there."
url: https://buildfirstbrain.com/journal/first-principles-thinking-is-graph-thinking/
canonical: https://buildfirstbrain.com/journal/first-principles-thinking-is-graph-thinking/
author: "Lawrence Arya"
authorUrl: https://www.linkedin.com/in/vibecoding/
published: 2026-06-07
updated: 2026-06-07
category: "Networked Thought"
tags: ["first principles", "thinking", "first brain", "networked thought", "problem solving"]
lang: en
---

# How to Think From First Principles: Find Root Nodes

> **TL;DR** Thinking from first principles is graph surgery: take an inherited conclusion, descend through repeated "how do we know that?" until you hit root nodes, facts and constraints that do not derive from anything else in the problem, then rebuild upward using only those. The descent sorts claims into physical constraints, verified data, conventions, and assumptions wearing fact costumes; the rebuild creates room for combinations the inherited assembly forbade. It is expensive, so reserve it for stuck problems, big bets, and suspicious claims, and let reasoning by analogy handle the routine, which is what conventions are for.

Think from first principles by treating it as graph surgery: take the conclusion you inherited, trace its supporting edges downward through repeated "how do we know that?", and keep descending until you hit **root nodes**, facts and constraints that do not derive from anything else in the problem. Then rebuild upward using only those. The method has three movements, deconstruct, verify, reconstruct, and one honest price tag: it is slow, which is why it is reserved for problems that are stuck, expensive, or suspicious, while analogy handles the routine. What you get for the price is a conclusion whose entire support structure you have personally inspected, which is a different kind of knowing than the pre-assembled beliefs most reasoning runs on.

## What is a first principle, in graph terms?

A node with no incoming edges inside the problem: nothing in the argument supports it because it needs no support from the argument, it is supported by physics, mathematics, direct measurement, or some human constant you can verify. The idea is ancient; Aristotle built his whole account of knowledge on [first principles as the primary basis from which a thing is known](https://plato.stanford.edu/entries/aristotle-metaphysics/), the foundations that cannot be derived from anything more basic. In your **biological knowledge graph**, these are the bedrock layer: batteries are made of atoms with market prices, humans read about 250 words a minute, the legal deadline is the legal deadline.

Everything above the bedrock is assembly, and here is the uncomfortable census: most of any mind's assembly was inherited, not built. Reasoning by analogy, "this is like that, so do what worked there", is the act of copying someone else's subgraph wholesale, which is efficient and usually fine, and quietly imports every assumption the original builder made for a different situation. First-principles thinking is the refusal to copy: it asks what the bedrock actually is, here, now, for this problem, and accepts only conclusions that can be rebuilt from it.

## How do you deconstruct a problem to its roots?

By descending one "why" at a time and sorting what you find. The descent technique is old enough to have a manufacturing pedigree: Toyota's [five whys](https://www.lean.org/lexicon-terms/5-whys/) drills from any symptom toward root cause by refusing to stop at the first plausible answer, and the same drill works on beliefs: why is this product priced at X, why does this process take six weeks, why do we "know" customers will not pay for that. Each answer gets the question again, until the answers stop deriving and start being checkable facts.

The canonical modern example, told well in [Farnam Street's treatment of first principles](https://fs.blog/first-principles/), is the battery-pack descent: the inherited node said batteries are expensive and always will be; the descent asked what batteries are made of and what those materials cost on the metals exchange, and found the bedrock price was a fraction of the assembled price, meaning the expense lived in assembly and convention, layers that engineering could attack. The descent did not produce the engineering; it revealed which node was actually load-bearing.

| What you find in the descent | Example | What to do with it |
| --- | --- | --- |
| Physical or mathematical constraint | Material costs, bandwidth limits, arithmetic | Keep; this is bedrock |
| Verified data | Measured churn, tested reading speed | Keep, with its error bars |
| Convention | "Invoices are net-30", "courses are 12 weeks" | Negotiable; ask what it protects first |
| Assumption in a fact costume | "Users won't pay for privacy" | Demote to hypothesis; test if it matters |
| Preference or identity claim | "We are not that kind of company" | Name it honestly; decide, don't derive |

## How do you rebuild from the roots?

Upward, with only verified nodes as foundation, and with deliberate openness about the edges. Reconstruction is where the method pays: once the problem is reduced to its bedrock, the combinations the old assembly forbade become visible, the roots can be wired in ways no inherited solution tried, and **insight as distant-node connection** stops being luck and becomes procedure. Most genuinely new solutions are old roots, new edges: the constraint set did not change; the assembly did.

Two disciplines keep the rebuild honest. Write the new chain out explicitly, claim by claim, so every edge from bedrock to conclusion is visible and challengeable, a one-page derivation, not a feeling of having thought hard. And mark the joints where you had to make a judgment call, because those joints are where your rebuild will fail if it fails, and knowing their location is most of debugging it later. A rebuilt conclusion with visible joints beats an inherited one with invisible everything, [which is the actual moat when machines can copy every assembly instantly](/journal/building-a-cognitive-moat-against-ai/).

## How do you train first-principles thinking?

With small daily descents and explanation from scratch. The training rep: once a day, take one thing you believe, a price, a best practice, a "fact" from your field, and descend three whys, sorting what you hit into bedrock, convention, assumption, and preference. Ten minutes, and within weeks the reflex changes: claims start arriving with an automatic "says who, supported by what?" attached.

The deeper rep is the [Feynman technique](https://fs.blog/feynman-technique/): explain the concept from scratch, in plain words, as if to a smart twelve-year-old, and watch precisely where the explanation breaks, because every break marks an inherited node you never actually built. This is also the cure for [tutorial hell](/journal/why-tutorial-hell-is-a-first-brain-failure/), where learners accumulate assemblies they cannot rebuild, and it transforms exam preparation: deriving a formula once from its roots beats rehearsing it fifty times, because the derivation installs the structure that regenerates the formula on demand, [the move that cracks competitive exams](/journal/the-first-brain-guide-to-cracking-competitive-exams/) at a fraction of the rote cost. Learners who map a field's root nodes first, then attach everything else to them, are doing [rapid skill acquisition by architecture](/journal/rapid-skill-acquisition-via-neural-mapping/) rather than by hours, and that graph-first construction habit is the core discipline of Building Your First Brain, free for the first 1,000 readers.

## When is first-principles thinking the wrong tool?

More often than its fans admit. The descent-and-rebuild cycle is expensive, hours to days of real work, and most decisions do not repay it: routine choices in familiar territory are exactly what conventions and analogies exist for, and a person who re-derives everything ships nothing. Conventions also encode survivors' wisdom: the fence in the field, per Chesterton's old warning, was usually built for a reason, and the descent should ask what a convention protects before celebrating its demolition, sometimes the answer is "thirty years of accidents you have not had yet."

The honest deployment rule: first principles for stuck problems where every analogy has failed, for big bets where being conventionally wrong is unaffordable, and for suspicious claims where someone profits from your not checking. Analogy for everything else, consciously, as a chosen efficiency rather than an unexamined default. And one boundary from the delegation age: an AI can run a plausible-sounding descent for you in seconds, but accepting its bedrock unverified just swaps one inherited assembly for another, faster one, [the outsourcing of the exact muscle the method exists to build](/journal/ai-agents-and-the-delegation-of-thought/). Use the machine to propose roots; verify them yourself, because the verification is the thinking.

## Key takeaways: thinking from first principles

First principles are root nodes: claims that survive the descent of repeated "how do we know?" because physics, math, or measurement holds them up. The method is three movements, descend and sort (constraint, data, convention, assumption, preference), verify the bedrock, rebuild upward with explicit, written edges and marked judgment joints. Train it with one three-why descent daily and Feynman explanations that expose inherited nodes. Spend it where it pays, stuck problems, big bets, suspicious claims, and let analogy serve the routine, after asking what each fence was protecting.

## Frequently asked questions

### How do you think from first principles?

Descend, verify, rebuild. Take the conclusion in front of you and ask "how do we know that?" repeatedly, sorting what you hit: physical constraints and verified data are bedrock; conventions are negotiable; assumptions in fact costumes get demoted to hypotheses. Then reconstruct the answer using only the bedrock, writing the chain out claim by claim so every link is visible. Reserve the full cycle for problems that are stuck, expensive, or suspicious; it is too slow for the routine.

### What is an example of first-principles thinking?

The battery descent: the inherited belief said battery packs are inherently expensive. Asking what batteries are physically made of, and what those raw materials cost on commodity markets, revealed bedrock far below the assembled price, which located the expense in assembly and convention rather than nature, and made the engineering attack obvious. The pattern generalizes: descend to what is physically or mathematically true, and see how much of the "impossible" was actually inherited.

### What is the difference between first principles and reasoning by analogy?

Analogy copies a working assembly: "this is like that, so do what worked there", fast, usually fine, and silently importing every assumption the original was built on. First principles rebuilds from bedrock: slower, costlier, and immune to inherited error, with room for combinations the copied assembly forbade. Strong thinkers use both deliberately: analogy as the default for routine territory, first principles where analogies keep failing or the stakes punish conventional wrongness.

### How do you practice first-principles thinking daily?

Two small reps. The three-why descent: once a day, take something you believe, a price, a process, a best practice, and ask why three times, sorting the answers into constraint, data, convention, or assumption. And the from-scratch explanation: pick one concept you use and explain it plainly without notes; every place the explanation breaks marks an inherited node you never built. Ten to fifteen minutes total, and the questioning reflex installs within weeks.

### When should you not use first-principles thinking?

For routine decisions in familiar territory, where conventions and analogies deliver ninety percent of the value at five percent of the cost, and where re-deriving everything becomes a way of never shipping. Also pause before demolishing a convention: many encode accumulated safety you cannot see, so ask what the fence protects first. The method earns its cost on stuck problems, irreversible bets, and claims whose seller benefits from your not checking.

## Dive deeper in

- [Building a Cognitive Moat Against AI](/journal/building-a-cognitive-moat-against-ai/)
- [Why Tutorial Hell Is a First Brain Failure](/journal/why-tutorial-hell-is-a-first-brain-failure/)
- [The First Brain Guide to Cracking Competitive Exams](/journal/the-first-brain-guide-to-cracking-competitive-exams/)
- [Rapid Skill Acquisition via Neural Mapping](/journal/rapid-skill-acquisition-via-neural-mapping/)

---

Source: https://buildfirstbrain.com/journal/first-principles-thinking-is-graph-thinking/
Author: Lawrence Arya — https://www.linkedin.com/in/vibecoding/