You’ve learned the moves in the abstract: models STACK when they point the same way, CHECK when one vetoes another’s blind spot, and COMPLETE when each explains a different slice of the same beast. Fine. But knowing the chords isn’t the same as playing a song.
So let’s play one. We’ll take a single, tense, real-world decision and run the whole lattice over it — several models firing at once — and watch a picture assemble that no single lens could have drawn. Then we’ll do the harder thing: figure out how you knew which lenses to grab in the first place, so the method transfers to situations you’ve never seen.
The goal of this lesson isn’t to memorise the bank-run answer. It’s to feel the procedure: point multiple models at one situation, tabulate what each sees, notice where they agree and where they fight, and let the disagreement do the thinking.
The case: a bank run
Here’s your situation. It’s a Tuesday. A rumour is going around — a friend-of-a-friend, a jittery post online — that your bank is “in trouble.” You walk past a branch and there’s a queue out the door, maybe forty people, shuffling, checking phones. Your entire savings are in that bank.
The question is brutally simple: do you pull your money out right now, or do you hold?
Your gut has an answer already. Ignore it. Instead, let’s run the lattice — deliberately pointing one model after another at the same queue — and see what each one sees.
Before you read — take a guess
Before we analyse it: what's the single most important reason a bank run can be dangerous even when the bank was basically fine that morning?
Lens 1 — Incentives: who gets rewarded?
Start where Munger says to start: follow the incentives. A bank doesn’t keep your deposit in a drawer with your name on it; it lends most of it out. So it can only pay back a fraction of depositors on any given day. That means there’s a queue-position prize: whoever withdraws first is safe; whoever withdraws last may find the till empty.
Read purely through incentives, the situation is grim. You are personally rewarded for running, and personally punished for hesitating. Lens 1 leans hard toward run.
Lens 2 — Game theory: what will everyone else do?
But your payoff doesn’t depend only on your move. It depends on everyone’s. This is game theory — the study of choices where the best thing for you to do hinges on what others do.
Model it as a coordination game. If almost nobody runs, the bank is fine and your money is safest sitting there earning interest — so you should hold. If almost everybody runs, the bank empties and holding gets you wiped out — so you should run. There are two self-consistent outcomes (“everybody holds” and “everybody runs”), and which one you land in is decided by expectations. “Everyone withdraws” is a self-fulfilling equilibrium: it’s stable precisely because, once you believe it, running becomes your best response.
That’s the trap. Your rational move flips entirely depending on a belief about strangers. Lens 2 doesn’t cleanly say run or hold — it says it depends on what you expect the crowd to expect. Which is exactly why the visible queue matters so much.
Lens 3 — Feedback loops: does it feed on itself?
Now ask the systems question: what’s flowing, and does the flow change the thing that produces it? This is a feedback loop.
Each withdrawal pulls cash out of the bank, which weakens the bank, which lengthens the queue, which frightens more people, which produces more withdrawals. Output loops back as input, amplified. That’s a reinforcing loop — the engine of every run, avalanche, and viral moment. Left alone, it doesn’t settle; it accelerates. Lens 3 leans toward run, and worse, warns that “run” gets more correct every minute you wait.
Lens 4 — Social proof: what is the queue saying?
Why is the queue itself such a powerful signal? Because of social proof — our habit of reading other people’s behaviour as evidence about what’s smart, especially under uncertainty. Forty people standing in line isn’t just forty people; it’s forty implicit testimonials that “sensible folks are getting their money out.” You didn’t verify the bank’s balance sheet. You outsourced the judgment to the queue. Lens 4 leans toward run — and notice it’s doing the same work as game theory (raising your estimate of how many others will run), which is why they STACK.
Lens 5 — Critical mass: is there a tipping point?
Does a little bit of running matter? Critical mass says: below a threshold, no; above it, catastrophically yes. A few nervous withdrawals get absorbed and the rumour fizzles. But past some tipping fraction, the reinforcing loop overwhelms the bank’s cash buffer and the run “goes critical” — it sweeps through the whole depositor base fast. Lens 5 reframes the decision as timing: the cost of running is low, the cost of holding is low — right up until the instant it isn’t, and there’s no gentle middle.
Lens 6 — Base rates: how often does this actually end in ruin?
Time for a CHECK. Base rates ask the boring, powerful question: of all the times a situation like this has occurred, how often did the bad outcome happen?
Most bank rumours are just rumours — they fizzle by Thursday. And in most modern systems, deposits up to a limit are insured, meaning even a genuine failure doesn’t cost you your covered savings. So the base rate of “I lose my insured money because I held” is low. Lens 6 leans firmly toward hold, and it’s directly contradicting the four lenses above it. Good — that contradiction is a feature.
Lens 7 — Second-order thinking: and then what?
The final CHECK. Second-order thinking asks what happens after the obvious first move — the consequence of the consequence.
The first-order logic is “pull my money out to be safe.” But run that forward: if everyone reasons “I’ll just pull mine out to be safe,” the collective withdrawal is precisely what drains the bank and causes the collapse nobody wanted. The move that’s individually protective is collectively ruinous. Second-order thinking doesn’t just lean toward hold — it exposes the whole situation as a tragedy: rational individuals producing a disaster none of them chose.
The multi-lens table
Here’s the whole reading in one grid. This table is the lattice for this problem:
| Lens | What it sees | Leans |
|---|---|---|
| Incentives | First to withdraw is safe; late withdrawers can lose | Run |
| Game theory | Coordination trap; “all run” is self-fulfilling | Run (if crowd runs) |
| Feedback loops | Each withdrawal weakens bank → more withdrawals | Run |
| Social proof | Visible queue signals “smart people are leaving” | Run |
| Critical mass | Past a tipping point the run sweeps fast | Run (timing) |
| Base rates | Most rumours fizzle; deposits usually insured | Hold |
| Second-order | If all run, the run causes the collapse | Hold |
Look at the shape. Five lenses STACK toward run — and that stacking is real; it’s exactly why bank runs happen and why they’re so hard to stop. But two lenses — base rates and second-order thinking — CHECK against them, pointing the other way for reasons that are also real.
The stacked “run” reading is not wrong — it correctly describes the local, individual incentives. But if you only ran the first five lenses, you’d conclude “obviously run” and miss that (a) the odds of actually losing insured money are low, and (b) your running is part of the machinery causing the collapse. A single-lens desk sprints confidently off a cliff.
The honest output isn’t a tidy verdict — it’s a flagged tension. The desk reports: strong local pressure to run, checked by a low base rate of true loss and a second-order warning that running is self-defeating at scale. That tension is the tragic core of the bank run: the individually rational move produces the collectively ruinous outcome. Whether you personally should run then depends on the details the CHECK forces you to actually look up — is your balance under the insurance limit? Is the rumour credible or vapor? — instead of just obeying the loudest lens.
The decision desk, driven by you
Reading a table is passive. Flip the lenses yourself. Below, each lens is a switch — turn it on and watch it pull the needle toward “Hold” or “Join the run.” Stack the run-lenses and you’ll feel the pressure build; add the CHECK lenses and watch the needle get honestly conflicted.
The decision desk
Should you join the bank run?
Switch on one model-lens at a time. Each lays its own reading on the desk and nudges the needle. Watch what happens when several agree — and what happens when they pull against each other.
A rumour says your bank is in trouble and a queue of forty people is forming outside the branch. Pull your savings now, or hold?
The desk readsNo lens is on yet. Turn one on to see what a single model says — then add more and watch them agree or clash.
The point of playing with it: there’s no toggle that makes the tension disappear. The lattice’s job isn’t to hand you comfort — it’s to make sure both the pressure to run and the reasons not to are on the table at the same time, so you decide with your eyes open.
The meta-skill: choosing which models fit
Here’s the question that separates people who own a latticework from people who’ve merely memorised one: for a brand-new situation you’ve never analysed, how do you know which lenses to grab?
You can’t run all two hundred models on every decision — that’s paralysis, not wisdom. But you don’t need to. A handful of interrogation questions reliably surfaces the models that matter. Point these at any situation and the relevant lenses light up:
- What DISCIPLINE does this live in? Physics, biology, economics, psychology? Name the field and you inherit its core models. A queue outside a bank is economics and psychology, so you’d expect incentives, base rates, and social proof to apply — and they did.
- WHO are the players? If two or more agents’ choices depend on each other, you’re in game theory territory. The bank run had depositors reacting to depositors — a dead giveaway for a coordination game.
- What is FLOWING or ACCUMULATING? Cash, users, heat, trust, anger. Anything that flows and loops back is a feedback loop or a stock-and-flow. Withdrawals draining a cash reserve while feeding more withdrawals — that’s the signature.
- What are the INCENTIVES? Who is rewarded for what? Follow the payoffs and the behaviour usually explains itself. “First out is safe” told us most of the story before any other lens spoke.
- What is the BASE RATE? Zoom out: of all times this pattern occurred, how did it usually end? This is your automatic CHECK against whatever the vivid, local lenses are screaming.
- What is the SECOND-ORDER effect? Play the obvious move forward one step. If everyone does it, what happens then? This is how you catch self-defeating strategies and tragedies-of-the-commons.
Notice these aren’t answers — they’re questions that summon the right models. That’s the whole trick. You don’t retrieve lenses by rummaging through a mental list; you retrieve them by interrogating the situation until the applicable ones announce themselves.
Circle of competence — and knowing when you’re lost
There’s a seventh question hiding under the other six, and it’s the most important one: can you even tell which lenses apply?
This is your circle of competence — the domain where you actually know enough to model well. Inside it, the interrogation questions get crisp answers. Outside it, you get static: you can’t name the discipline, you don’t know who the real players are, you can’t spot the base rate because you’ve never seen the reference class.
That static is information, not embarrassment. When you genuinely can’t tell which lenses fit, the correct move is not to bluff a latticework — it’s to notice you’re out of your depth and act accordingly: defer, get a specialist, or bound your bet so being wrong doesn’t ruin you.
Confidently running the wrong lenses is more dangerous than running none, because the analysis feels rigorous. A slick multi-model story built on a domain you don’t understand is just a well-decorated guess. “I can’t tell which models apply here” is a sophisticated conclusion, not a failure of nerve.
When to use it
Reach for the full interrogation-and-tabulate procedure when a decision is (a) important enough to be worth ten minutes of structured thought and (b) tangled enough that one obvious lens might be missing something. Everyday, reversible, low-stakes choices don’t need the whole desk — a single good model, or just moving on, is fine. Save the lattice for the decisions where being wrong is expensive and where the first lens that pops into your head is probably not the only one that matters.
A second, transfer mini-case
A method that only works on the example it was taught with is a party trick. Let’s prove the procedure transfers by pointing it — quickly — at a completely different situation.
The case: you’re offered a job at an early-stage startup. Lower salary than your current gig, a pile of stock options, a founder who’s magnetic in the interview, and a product that’s “about to explode.” Take it or stay put?
Run the interrogation questions and the lenses fall out:
| Lens | What it sees | Leans |
|---|---|---|
| Incentives | The founder is incentivised to sell you the dream; your interviewer’s job is recruiting, not disclosure | Discount the pitch |
| Base rates | Most early-stage startups fail; most option grants end up worth zero | Stay / discount |
| Asymmetry & optionality | Downside is capped (you can leave; salary is a floor); upside if it hits is huge | Lean take |
| Second-order thinking | If it fails, what’s the résumé and network worth afterward? Sometimes a lot | Lean take |
| Social proof | ”Everyone great is joining” may be manufactured hype, not signal | Discount |
Same procedure, entirely different content — and again the lenses don’t agree, which is the whole value. Base rates and skepticism about the pitch CHECK the founder’s stacked optimism; asymmetry and second-order effects reveal that the downside is smaller than it feels while the upside is real. The lattice doesn’t tell you to take the job. It tells you the honest decision is about your appetite for a capped-downside, low-base-rate, high-upside bet — which is a far better question than “is the founder inspiring?”
Recap: running and building a lattice
In the bank-run analysis, five lenses leaned toward 'run' and two toward 'hold'. What's the right way to describe that output?
Check your answer to continue.
Where this goes next
You can now run the lattice on a live decision and choose which lenses to run — the two halves of the skill. But a procedure this powerful has its own failure modes: it can be faked, over-applied, and turned into a tool for rationalising whatever you already wanted.
In the next lesson, The Checklist — and Where the Lattice Lies, we’ll compress everything into a portable checklist you can actually use under pressure — and then, crucially, examine where the latticework misleads you: motivated model-picking, false precision, and the seductive multi-model story that’s confidently, elegantly wrong.