By now you own the big idea: a latticework beats a single man-with-a-hammer view because you run several models over one situation at once. Lovely. But “run several models” is a slogan, not a technique. What actually happens when two models meet over the same problem? Do they add? Argue? Split the work?
It turns out there are exactly three moves, and they feel completely different from the inside. Two models can stack (both point the same way and their forces multiply), check (one catches the blind spot of the other), or complete (each explains a different slice of the same thing). Same lattice, three verbs. Get the verb wrong and you’ll trust a warning as if it were proof, or throw away a veto because it “disagreed with” your favorite lens.
This lesson is the engine room of the whole skill. Learn to feel which of the three is happening, because — spoiler — your response should differ each time.
Before you read — take a guess
Four independent models all point at the same conclusion — 'buy now'. What's the most useful way to read that pile-up?
Stacking — when models point the same way
Analogy. Picture four people leaning on the same stuck door. Each shove alone barely moves it. Together the door doesn’t just open a little more — it bursts. The forces don’t line up politely side by side; they compound. That’s stacking.
Precise definition. Stacking is when several independent models all lean in the same direction on the same decision, and their forces combine multiplicatively rather than additively. When the product of many same-direction forces blows past what any of them could do alone, you get a lollapalooza effect — the exact phenomenon from the lollapalooza-effect course: outcomes that look impossibly extreme until you notice how many models were secretly rowing the same way. Stacking is the mechanism; the lollapalooza is its result.
Two engines make the multiplication real:
- Feedback loops. Each model’s push changes the situation so the next model pushes harder. Rising prices are social proof; social proof draws buyers; more buyers raise prices. The loop feeds itself.
- Critical mass. Below a threshold the stacked forces fizzle; above it they sweep the whole system. A few excited buyers do nothing; enough of them and the mania tips and runs on its own.
Worked example: a market mania
Say a hot stock is climbing and you feel the pull to jump in. Instead of asking “is it going up?” (one model), lay every model that’s acting on you side by side:
| Model | What it’s whispering | Direction |
|---|---|---|
| Incentives | ”Everyone around you got rich; you’re the sucker who didn’t” | Buy |
| Social proof | ”All these smart-seeming people are buying — they must know something” | Buy |
| Commitment | ”You already told friends you’re bullish; backing out means eating your words” | Buy |
| Scarcity | ”Only a few shares left at this price, act NOW before it’s gone” | Buy |
Four independent forces, one direction. Additively that’s “four reasons to buy.” Multiplicatively it’s a force that overrides your judgment entirely — which is precisely how manias, cults, and disastrous group decisions get built. The stack didn’t make the stock a good buy. It made not buying feel almost physically impossible.
A stacked reading is a strong signal and a red flag at the same time. When every model you own points one way, that can mean the situation is genuinely lopsided — or that you’ve been captured by a lollapalooza and can no longer see straight. The correct move on a big stack is never “so, obviously, go.” It’s “so, obviously, slow down and check.” Which is a nice segue.
When to use it
Reach for the stacking read whenever you notice many models agreeing, especially about a high-stakes, emotionally charged choice. The agreement is information — but the strength of the agreement is a separate signal telling you to add friction: sleep on it, seek a disconfirming view, size the bet down. Stacking earns conviction and demands caution in the same breath.
Checking — when models veto each other’s blind spots
Analogy. A single witness tells a gripping story and you believe every word — until a second witness, who was standing at a different angle, says “that’s not what happened.” The collision is uncomfortable, and the collision is exactly what saves you from the confident-but-wrong first account.
Precise definition. Checking is when one model catches an error that another model can’t see from where it stands. The models don’t agree; one vetoes the other. Because every model has a blind spot — a class of situation where it’s confidently wrong — the point of a second model is often not to add force but to catch the first one lying. When two good models collide, that collision is data.
Worked examples: three classic vetoes
| The seductive model | The model that vetoes it | Who wins, and why |
|---|---|---|
| A vivid, specific story (“this founder is a visionary, look at the demo!”) | The cold base rate (“~90% of startups like this fail”) | The base rate wins. A great story feels like evidence; it’s mostly narrative. Anchor to the reference class first, adjust for the story second. |
| The stated reason (“we’re doing this for the customer”) | The actual incentive (“whoever pushes this gets promoted”) | Follow the incentive. When the reason someone gives and the reward they’ll collect point different ways, the reward is the better predictor of behavior. |
| The appealing first-order move (“just ban it — problem solved”) | Its second-order consequence (“a black market forms and it gets worse”) | Play the tape forward. The obvious fix vetoes itself once you ask “and then what?” — the domain of second-order thinking. |
Notice the pattern: the appealing model is the one your brain wants to obey, and the checking model is the annoying one that ruins the vibe. That annoyance is the sound of a blind spot being covered.
When two of your models flatly disagree, resist the urge to average them into a mushy compromise. Ask which one is operating outside its competence here. A base rate beats a story about base-rate-type questions; a story might beat a base rate when you genuinely have private, reliable information. The veto is only worth trusting when the vetoing model is the one on home turf.
When to use it
Run a deliberate check whenever a conclusion feels too clean, too emotionally satisfying, or too convenient for the person proposing it. Pick the model most likely to embarrass your favorite one — base rate against story, incentive against stated reason, second-order against first-order — and let them collide on purpose. If the appealing model survives an honest veto attempt, now you can trust it.
Completing — when each model explains a different part
Analogy. Three doctors examine the same patient. The cardiologist reads the heart, the neurologist reads the brain, the radiologist reads the scan. They’re not arguing and they’re not piling on — each one describes a different organ, and only together do they describe the whole body.
Precise definition. Completing is when several non-competing lenses each explain a different part of one situation. They don’t stack (they’re not pushing the same lever) and they don’t check (they’re not correcting each other) — they divide the labor. Each model owns the slice of reality it was built for, and you assemble the slices into a picture no single lens could draw.
Worked example: reading a housing bubble
Ask three different questions about the same overheated housing market and hand each to the model built to answer it:
| Lens | The part it explains | What it says here |
|---|---|---|
| Supply and demand | The price | Cheap credit and limited new building push prices up; there’s the number on the sticker. |
| Game theory | The players | Buyers fear being priced out forever, so they bid over ask; lenders compete to approve. Everyone’s rational move worsens the whole. |
| Feedback loops | The dynamics | Rising prices become collateral for bigger loans, which fund higher bids, which raise prices — the self-reinforcing spiral that turns a rise into a bubble. |
No lens is wrong; none is fighting the others. Supply-and-demand can’t tell you why buyers panic-bid — that’s the players’ game. Game theory can’t tell you why it accelerates — that’s the loop. Stack all three and you don’t get a bigger number, you get a fuller one: the price, the psychology, and the motion, at once.
The same trick reads a traffic jam. Supply and demand: too many cars for the road (the price of space, paid in time). Game theory: each driver merges late and brakes for themselves, worsening it for all — a tragedy of the commons on wheels. Feedback loops: one brake tap ripples backward into a standing wave that persists long after the original cause is gone. One jam, three complementary explanations, each true.
Completing is the quiet workhorse of the latticework. It rarely produces the drama of a stacked lollapalooza or the save of a veto — it just steadily makes your picture whole instead of a keyhole view. Most real analysis is mostly completing, punctuated by the occasional stack or check.
When to use it
Default to completing on any rich, multi-sided situation — a market, an organization, a policy, a relationship. Ask: what are the different parts of this thing, and which model owns each part? Prices and quantities → supply and demand. Strategic actors → game theory. Change over time → feedback loops. Assign the lenses to the slices, then read them together.
Which of the three is happening?
Here’s why naming the verb matters: your correct response is different for each. Mistake a check for a stack and you’ll pile “confidence” onto a conclusion that another model was trying to veto. Mistake completing for stacking and you’ll invent a fake lollapalooza out of models that were just describing different organs.
| If the models… | …you’re seeing | …and the right move is |
|---|---|---|
| Point the same way, forces multiply | Stacking | Conviction and caution — take it seriously, then deliberately slow down and check |
| Disagree, one catches the other’s error | Checking | Trust the veto (from the model on home turf); don’t average it away |
| Point at different parts of the situation | Completing | Assemble the fuller picture; don’t mistake breadth for agreement |
A quick diagnostic: ask, “are these models pushing the same lever, correcting each other, or describing different pieces?” Pushing the same lever → stacking. Correcting → checking. Describing different pieces → completing. Same three models over the same case can even do all three depending on the question you ask them — which is the whole reason the lattice is worth building.
Try it: the decision desk
Below is a real-ish call: your team wants to cut prices to beat a rival. Flip lenses on and off and watch the needle. Turn on the lenses that agree and you’ll produce a stacked reading (the needle swings hard one way). Turn on lenses that pull opposite ways and you’ll surface a tension to check. Every lens on at once is the complete picture — messier, and truer.
The decision desk
Decision desk: cut prices to beat the rival?
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 rival just undercut you. Your team wants to slash prices to win the quarter. Hold or cut?
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.
Notice what the desk teaches. The lenses that shout “cut” — first-order, social proof, sales incentives — stack into a strong, urgent pull. But flip on second-order and game theory and they veto the whole thing: the appealing move is a trap. And read all of them together and you complete the picture — you understand why the pressure to cut is real, why cutting is still probably wrong, and exactly which force to distrust (the sales team’s incentive). Three verbs, one desk.
Sort the moves
Each statement below is a small example of one combination move. Drop it in the right bucket.
Sort each statement into the combination move it describes: models pointing the same way (Stacking), one model catching another's blind spot (Checking), or each model explaining a different part (Completing).
- To read a traffic jam you use crowding for the cause, driver strategy for the behavior, and ripple loops for the standing wave.
- A thrilling founder story runs into a brutal base rate, and the base rate wins.
- Commitment, authority, and reciprocity all push the same direction until the group tips past the point of no return.
- Rising prices are social proof, which draws buyers, which raises prices further — a self-feeding pile-up.
- Incentives, social proof, and scarcity all scream “buy now” and the urge becomes overwhelming.
- Supply and demand set the price, game theory explains the bidders, feedback loops explain the spiral.
- The obvious “just ban it” fix meets its second-order consequence — a black market — and gets overruled.
- The stated reason (“for the customer”) collides with the real incentive (a promotion), so you follow the incentive.
Recap quiz
Three ways models combine
What distinguishes stacking from mere addition of reasons?
Check your answer to continue.
Where this goes next
You now have the three verbs of the lattice — stack, check, complete — and a feel for how the right response changes with each. The next lesson, Running the Lattice: A Case Study, walks one real situation end to end, deciding at every turn which of the three moves is in play and letting the full latticework do its work. Bring the three verbs; you’re about to use all of them at once.