α
AlphaSteve
← Brain index

Mental Models — The Latticework

Why this exists: see 00-mission — the latticework is how breadth of inputs gets compressed into edge.

Charlie Munger's central idea: you reason better by drawing on a small number of fundamental models from many disciplines than by mastering one. This file is the agent's working latticework. None of these is the analysis; all of them are checks against tunnel vision.

The hard core — the models you reach for first

1. Margin of safety (engineering / finance)

A bridge rated for 30 tons should carry 10. A thesis that requires every assumption to hit is a thesis with no safety. Always ask: what is the safety factor on each input?

2. Circle of competence (epistemology)

Know what you know and, more importantly, know the edge of it. The edge is the dangerous place. Inside the circle: act. Outside: pass. On the edge: do more work or pass. Saying "I don't know" is the most underrated value-creating move in this work.

3. Inversion (mathematics, via Jacobi: "invert, always invert")

Don't ask "how do I make money on this?" Ask "how do I lose money on this?" Build the bear case first. If you cannot articulate the bear case more convincingly than the bulls, you have not understood the situation. See 04-Market-Read/pre-mortem-and-inversion.md.

4. Second-level thinking (Howard Marks)

First-level: "the business is good, I should buy it." Second-level: "the business is good, everyone agrees, so the price reflects that — is there a reason to disagree on either fact or interpretation?" Edge comes from second level. See 04-Market-Read/second-level-thinking.md.

5. Base rates (statistics)

What is the historical base rate of: turnarounds succeeding, M&A deals creating value, "this time is different" being correct, a founder-CEO remaining capable post-IPO? The "outside view" is usually closer to right than the inside-view story. See 06-Risk/base-rates.md.

6. Expected value (probability)

A 30% chance of doubling and a 70% chance of losing 20% is a positive expected value position; people refuse it because it "loses most of the time." Conversely, a 95% chance of making 5% and a 5% chance of going to zero is a negative EV position that feels safe. Always think in distributions, not point estimates.

7. Bayesian updating (probability)

You have a prior. New information should update it — but only proportionally to its strength and independence. Most "news" is noise. Strong updates come from disconfirming evidence, primary data, and shifts in the underlying business — not from price movements.

The structural models — how businesses actually work

8. Capital cycle (Marathon / Edward Chancellor)

Capital flows toward returns, increasing supply, which depresses returns, which causes capital to retreat. Watch capex relative to depreciation across an industry; watch IPO and equity issuance windows. The capital cycle is why cyclicals are mispriced at turns. See 05-Environment/capital-cycle.md.

9. Porter's Five Forces (industrial organization)

Customer power, supplier power, threat of entry, threat of substitutes, competitive rivalry. Use it as a sanity check on a moat thesis, not as a checklist for grading. See 08-Frameworks/porters-five-forces.md.

10. The seven powers (Hamilton Helmer)

Network economies, scale economies, switching costs, cornered resource, branding, counter-positioning, process power. Each has a specific structural mechanism for why a competitor cannot copy. The single most useful schema for moat work. See 02-Business-Quality/Moats/moat-taxonomy-and-identification.md.

11. Unit economics (microeconomics)

Strip the business to its irreducible economic unit (a store, a customer, a route, a well, a subscriber) and ask: does this unit cover its variable cost, contribution margin, allocated overhead, and capital charge? If the unit does not make money, the company does not, regardless of scale. See 08-Frameworks/unit-economics.md.

12. Operating leverage and fixed cost absorption (cost accounting)

High fixed-cost businesses (utilities, semis fabs, airlines, railways) have explosive earnings sensitivity to small revenue changes. The same property that makes them dangerous on the way down makes them mispriced at troughs.

13. The S-curve / diffusion of innovations (technology adoption)

Adoption is rarely linear. Early flat → steep middle → saturation. Most "TAM is huge" narratives ignore where the curve is. Most cyclical disappointments ignore that the curve has bent.

14. Network effects (network science)

Real network effects compound non-linearly with adoption; they are not the same as "we have a lot of users." Direct (each user makes the product better for every other user), indirect (more users → more developers → more apps), local (regional), and data (more usage → better algorithm). See 02-Business-Quality/Moats/network-effects.md.

The financial models — how the numbers behave

15. ROIC vs. cost of capital

Value is created only when returns on invested capital exceed the cost of that capital. Growth at returns below cost of capital destroys value (this is counterintuitive but iron). All capital allocation analysis sits on this. See 02-Business-Quality/Fundamentals/roic-decomposition.md.

16. DuPont decomposition

ROE = Net Margin × Asset Turnover × Leverage. Decomposing ROE tells you why it's high. Margin businesses, turnover businesses, and leveraged businesses each have different sources of risk. See 02-Business-Quality/Fundamentals/roic-decomposition.md.

17. Cash conversion

Earnings without cash are usually fiction. Free cash flow / net income over a cycle is the single most informative quality-of-earnings ratio. See 02-Business-Quality/Fundamentals/quality-of-earnings.md.

18. The wave / cycle

Industries cycle: capex → overcapacity → price collapse → underinvestment → tightness → capex. Commodities, semis, shipping, refining, ad markets, real estate. Knowing where you are in the cycle is half the work in cyclical sectors.

19. Reinvestment runway

A 20% ROIC business that can reinvest only 10% of its earnings is worth far less than a 15% ROIC business that can reinvest 80%. Quality is not just ROIC; it is ROIC times reinvestment rate, sustained.

The behavioral models — how investors get fooled

20. Anchoring

Old prices, IPO prices, prior valuations, "what I paid" — all are anchors that contaminate fresh analysis. Ask: would I buy this today if I had never seen the chart?

21. Confirmation bias

You will find what you go looking for. Active inversion (model #3) is the disciplined antidote.

22. Narrative fallacy / story stocks

Coherent narratives feel like understanding. They are not. A story that explains everything explains nothing. Test stories against base rates.

23. Recency bias

Investors extrapolate the last 18 months. The capital cycle (#8) operates on a longer wavelength than that. Be skeptical of any thesis that requires the recent past to continue.

24. Authority bias

Famous investors are not always right; sell-side analysts have institutional incentives that diverge from yours; CEOs are paid to be optimistic. Source-rank everything (see 04-Market-Read/cognitive-bias-checklist.md).

25. Sunk cost fallacy

The fact that you've done weeks of work on a name does not create any obligation to have a view, much less a buy view. Walking away is sometimes the highest-value output.

The physics models — how things actually scale

26. Backwards induction (game theory)

Start from the endgame and reason backwards. What does this industry look like in steady state? Who wins? Who is forced out? What is the price of the marginal unit when the dust settles? Most cyclicals are best valued this way.

27. The law of large numbers / regression to the mean

Extreme returns regress. So do extreme deficits. Persistent outliers — businesses that earn very high or very low returns for very long — require a structural explanation (a moat or a curse). Without one, expect regression.

28. Convexity / asymmetry (options theory)

The shape of the payoff matters more than the expected value. Positions where downside is bounded (an asset floor, a hidden asset, a cash-rich balance sheet) and upside is uncertain are the deep-value sweet spot.

29. Lollapalooza (Munger)

When several mental models point the same direction, the conviction is non-linear. When models conflict, slow down. The Lollapalooza is also a warning: in manias, many biases stack up to produce wild mispricings. Look for inverse Lollapaloozas: where multiple biases stack against a fundamentally sound business.

How to use the latticework

Three rules:

  1. Reach for at least three models per analysis. A single-model view is a brittle view.
  2. Note when models conflict. Conflict is information — it tells you where to do more work.
  3. Do not force a model where it doesn't fit. Mental models are tools, not religion. If none of these illuminates the situation, say so and reach for primary research.

Linked memory