The First Principles Handbook

When a proxy becomes an optimization target, pressure can break its relationship with the underlying objective.

Use it whenever a KPI starts becoming more important than the reality it was meant to represent.

The more a quantitative indicator is used for social decision-making, the more pressure it faces for corruption and the more it can distort the process it measures.

Treat it as Goodhart’s organizational cousin: incentives can corrupt both metrics and behavior.

Prefer explanations that require fewer unnecessary assumptions, all else equal. Simpler does not mean simplistic.

Remove explanatory machinery until further removal hurts predictive power.

Before removing an apparently pointless rule, process or constraint, understand the problem it was originally solving.

The antidote to reckless “zero-based” redesign.

Many systems exhibit highly unequal distributions of causes and effects. The famous 80/20 ratio is a heuristic, not a constant of nature.

Search for the few causes responsible for a disproportionate share of the outcome.

Work tends to expand to fill the time made available for it.

Use constraints deliberately. More time can produce more polish, but also more scope, delay and ceremony.

Complex work often takes longer than expected, even when people attempt to correct for prior underestimation.

Plan with ranges, buffers and explicit uncertainty rather than single-point confidence.

Adding people to a late software project can make it later because onboarding and communication consume scarce capacity.

Headcount is not throughput. Coordination cost is itself a load on the system.

Choice reaction time tends to increase with the information or uncertainty contained in the set of alternatives.

Reduce needless options when speed and clarity matter; preserve options when exploration matters more.

Hierarchies can promote people based on performance in one role into a different role where the same competencies no longer transfer.

Promote against the requirements of the destination role, not merely the achievements of the current one.

Do not jump to hostile intent when error, ignorance, misaligned incentives or ordinary system failure can explain the outcome.

Use it to generate less dramatic hypotheses, then test the causal mechanism rather than assuming any one motive.

The original research found that low performers in several tasks substantially overestimated their relative performance. Popular retellings often overstate the shape and universality of the effect.

Calibrate confidence against external evidence, not felt fluency.

Past, unrecoverable costs should not determine whether the next unit of investment is worthwhile.

Ask: if we did not already own this project, would we choose to buy it today at its remaining cost?

Extreme observations are often followed by less extreme observations even without intervention, because luck and noise fluctuate.

Do not mistake natural reversion for the effect of your latest management action.

Before treating a case as unique, look at what usually happens in the relevant reference class.

First principles prevent blind imitation; base rates prevent exceptionalism from becoming fantasy.

Instead of asking only how to succeed, ask what would reliably cause failure, then remove or avoid those conditions.

Often easier to identify catastrophic paths than to specify an optimal one.

Beyond some point, additional input can produce progressively smaller incremental output while other factors are held fixed.

Ask where the next dollar or hour has stopped being the best use of the resource.

When one party acts on behalf of another but has different incentives or information, actions can drift away from the owner’s objective.

Design incentives and information flows so local decisions serve system-level intent.

Markets and organizations both incur costs of search, contracting, coordination, monitoring and enforcement.

The cheapest sticker price can be the most expensive system once friction is counted.

Past choices can alter the cost and feasibility of future choices. Current structure may be contingent rather than optimal.

Ask whether today’s constraint is fundamental or merely the accumulated consequence of yesterday’s decisions.

In some domains, a small number of outcomes account for most of the total value. Venture returns, audience size and network connectivity often behave more like heavy-tailed distributions than neat averages.

When upside is highly skewed, portfolio logic and prioritization should reflect that asymmetry.

For some non-perishable things, long survival can be weak evidence of continued survival. It is a heuristic, not a law, and selection effects matter.

Give durable ideas some prior credit, but never let age substitute for mechanism.