Latticework

Thinking

First principles

Reduce a problem to fundamental constraints and rebuild upward from there.

First principles are the most basic truths or assumptions underlying a subject — ideas you cannot derive from simpler ideas within that system.

First-principles thinking means solving a problem by:

Breaking it down into facts that are justifiable.

Separating those facts from convention, analogy, and “how it’s normally done.”

Building a solution upward from the fundamentals.

Reason from what must be true, not from what has usually been done.

Bayesian updating

Revise beliefs in proportion to new evidence.

Bayesian updating is adjusting confidence in a belief by the strength of new evidence, starting from what you already believed.

Updating means revising a belief by:

Stating your prior — how confident you were before the evidence arrived.

Asking how much more likely that evidence is if the belief is true than if it is false.

Moving your confidence by that amount, and no further.

Change your mind in increments proportional to the evidence, not in leaps proportional to the surprise.

Probabilistic thinking

Think in likelihoods and ranges rather than certainties.

Probabilistic thinking treats the future as a distribution of outcomes with weights, not as a single prediction that turns out right or wrong.

Thinking probabilistically means forming a view by:

Listing the plausible outcomes instead of the one you expect.

Assigning each a rough likelihood, and a magnitude if it occurs.

Choosing the action with the best weighted payoff, not the most likely story.

A good decision can still produce a bad outcome. Judge the process, not the draw.

Expected value

Weigh an outcome by both its payoff and its probability.

Expected value is the average result of a choice repeated many times — each possible payoff weighted by its probability.

Thinking in expected value means comparing options by:

Listing the possible outcomes and the payoff attached to each.

Weighting every payoff by its probability and summing them.

Taking the higher expected value, unless a loss along the way would end the game.

Take positive expected value repeatedly, and never at a size that can ruin you.

Base rates

Start with historical frequency before focusing on vivid specifics.

A base rate is how often an outcome occurs across the whole reference class, before you know anything about the specific case in front of you.

Reasoning from base rates means estimating by:

Naming the reference class the case actually belongs to.

Starting from that class's historical frequency rather than the details of the case.

Adjusting for the specifics only as far as they are genuinely diagnostic.

The vivid particulars of a case rarely justify abandoning what usually happens.

Inversion

Ask how to cause failure, then avoid those causes.

Inversion is attacking a problem from its opposite end — asking what would guarantee failure instead of what might produce success.

Inverted thinking means approaching a goal by:

Stating the outcome you want, then describing its exact opposite.

Listing everything that would reliably produce that failure.

Removing those things first, before adding anything clever.

Eliminating the causes of failure is easier than engineering the causes of success.

Second-order thinking

Consider the consequences of the consequences.

Second-order effects are the consequences of the consequences — what happens after the obvious first effect has played out.

Second-order thinking means evaluating a decision by:

Naming the immediate, first-order effect.

Asking “and then what?” at least twice more.

Weighing the later effects, which often reverse the sign of the first.

The first consequence is rarely the one that decides the outcome.

Systems thinking

Analyze interdependent parts, feedback, constraints, and delays.

Systems thinking explains outcomes from the structure connecting the parts — stocks, flows, delays, and loops — rather than from the parts themselves.

Thinking in systems means diagnosing a problem by:

Mapping the elements, their connections, and the purpose the system actually serves.

Looking for the delays and loops that make cause and effect look unrelated.

Intervening at structure and leverage points instead of at events.

Persistent problems are produced by structure, not by the people inside it.

Opportunity cost

Account for the best alternative forgone.

Opportunity cost is the value of the best alternative you gave up in order to do what you chose instead.

Thinking in opportunity cost means judging a choice by:

Identifying the single best alternative use of the same time, money, or attention.

Comparing the option in front of you against that alternative, not against nothing.

Accepting it only if it beats what you are forfeiting.

The real cost of anything is what you can no longer do because of it.

Circle of competence

Know the boundary of what you genuinely understand.

Your circle of competence is the set of domains where your judgment is reliably better than the average participant's.

Working within the circle means:

Defining honestly what you understand well enough to predict.

Marking the boundary and treating everything outside it as unknown.

Expanding the circle deliberately, rather than pretending it is already larger.

The size of the circle matters far less than knowing where its edge is.

The map is not the territory

Distinguish models, metrics, labels, and narratives from reality.

A map is a compressed representation of reality — useful precisely because it discards detail, and wrong for the same reason.

Holding a model correctly means:

Knowing what the model deliberately leaves out.

Checking the abstraction against the terrain whenever the two disagree.

Redrawing the map when reality moves, instead of defending it.

When the map and the terrain disagree, trust the terrain.

Goodhart's law

When a measure becomes a target, it stops being a good measure.

Goodhart's law holds that a measure stops being a good measure once it becomes the target people are judged on.

Guarding against Goodhart means designing metrics by:

Distinguishing the outcome you want from the proxy you can count.

Anticipating how the proxy can be raised without producing the outcome.

Pairing each metric with a counter-metric, or changing what you measure over time.

You get what you measure, including every way it can be produced without the thing you wanted.

Decisions

Optionality

Favor choices that preserve or create valuable future choices.

Optionality is holding the right to act without the obligation to — a position with a capped downside and an open-ended upside.

Building optionality means positioning by:

Paying a small, known cost to keep a large, unknown payoff available.

Refusing commitments that cap the upside while leaving the downside open.

Exercising only once the payoff is clear, and letting the rest expire.

Pay small to stay in the game, and let the asymmetry do the rest.

Margin of safety

Build room for error between assumption and failure.

A margin of safety is the buffer between what a system can withstand and what you expect to demand of it.

Designing with a margin of safety means:

Estimating the load, cost, or downside you expect.

Assuming that estimate is wrong in the unfavorable direction.

Sizing the buffer so being wrong is survivable rather than fatal.

Engineer for the error in the estimate, not for the estimate.

Antifragility

Gain from variability, stress, and disorder within limits.

Antifragility is the property of systems that improve under stress, volatility, and error — rather than merely resisting them, as robust systems do.

Building antifragility means structuring a system by:

Distinguishing fragile, robust, and antifragile responses to the same shock.

Capping the downside of each stressor so failures stay small and survivable.

Exposing the system to frequent small shocks so it adapts before a large one arrives.

The robust survives volatility. The antifragile requires it.

Compounding

Small repeated gains or losses accumulate nonlinearly over time.

Compounding is growth applied to the results of earlier growth — capital, skill, reputation, or relationships building on their own accumulated base.

Thinking in compounding means building by:

Choosing processes whose output feeds back into their own input.

Protecting the base, since interruptions and drawdowns cost more than they appear to.

Extending the horizon, since duration matters more than rate.

Rate gets the attention. Time does the work.

Probability

Regression to the mean

Extreme results tend to be followed by less extreme ones.

Regression to the mean is the tendency of extreme results, which mix skill with luck, to be followed by ordinary ones.

Accounting for regression means interpreting results by:

Separating the repeatable component of a result from the lucky one.

Expecting the next observation to sit closer to the long-run average.

Asking whether an intervention beat that expectation, or merely preceded it.

The outlier was mostly luck, and luck does not repeat on schedule.

Psychology

Sunk-cost fallacy

Continue because of unrecoverable past investment.

A sunk cost is time, money, or effort already spent and unrecoverable. The fallacy is letting it justify spending more.

Escaping a sunk cost means judging a commitment by:

Asking what you would choose today if you were starting fresh.

Counting only future costs against future benefits.

Treating what you already spent as information, not as a reason to continue.

The only question is whether the next dollar earns its return, never what the last one cost.

Economics

Network effects

The product becomes more valuable as more people use it.

A network effect exists when each additional user makes the product more valuable to every existing user.

Assessing a network effect means asking:

Whether new users add value to existing users, or only add revenue.

Whether the effect is global or local — the whole network, or one cluster inside it.

Where the effect turns negative through congestion, noise, or falling quality.

Growth is not a network effect unless users make the product better for each other.

Incentives

Rewards and penalties shape behavior more reliably than stated values.

An incentive is the payoff — money, status, safety, convenience — that makes a behavior rational for the person doing it.

Incentive analysis means explaining behavior by:

Identifying who is actually paid, promoted, or protected by the current outcome.

Assuming people respond to their own payoff, not to the stated goal.

Changing the payoff before trying to change the behavior.

Behavior that looks irrational usually makes sense once you find who is being paid.

Systems

Feedback loops

Outputs become future inputs.

A feedback loop exists when a system's output returns as its own input — amplifying the movement (reinforcing) or damping it (balancing).

Reading a system's loops means:

Tracing where an output comes back around as an input.

Determining whether that return amplifies or corrects the original movement.

Changing the loop's strength or delay instead of fighting its symptoms.

Behavior you cannot explain from the parts is usually explained by the loop.

Bottlenecks

Throughput is set by the slowest or scarcest step.

A bottleneck is the single constraint that sets throughput for an entire system, however much spare capacity exists everywhere else.

Working the bottleneck means improving a system by:

Locating the stage where work accumulates and waits.

Directing effort there, and only there.

Finding the constraint again after it moves, because relieving one creates another.

Improvement anywhere but the constraint is an illusion of progress.

Society

Skin in the game

Decision-makers should share the consequences of their decisions.

Skin in the game means bearing a real share of the losses from your own decisions, not only the gains.

Applying it means judging decisions and advice by:

Asking what the decision-maker loses if they turn out to be wrong.

Discounting confidence that costs its holder nothing.

Structuring arrangements so that authority and exposure sit with the same person.

Discount every opinion by what it would cost its owner to be wrong.

Using them

Take a real decision — whether to build a feature, say — and run a short checklist rather than the whole list.

First principles: what user problem and what constraint are actually present?

Jobs to be done: what progress is the customer hiring this to make?

Opportunity cost: what do we not build, learn, or fix if we build this?

Incentives: what behavior does this reward, for users, the team, and the business?

Second-order effects: what happens after adoption, abuse, support load, and the competitive response?

Bottleneck: is this the actual constraint on retention or conversion, or a comfortable adjacent one?

Goodhart's law: which metric improves while user value gets worse?

Optionality: can a prototype or staged release come before the irreversible commitment?

Expected value: what are the probability-weighted upside, downside, cost, and learning value?

Premortem: it failed six months from now. Why?

The point is not to apply every model. It is to hold a few competing lenses at once, because the reliable failure is not using the wrong model — it is using only one.