Ideas are easier to test once their machinery is visible.

These essays borrow models from machine learning—search, optimization, distributed updates, compounding—and use them to examine social systems. Each one is an argument about a mechanism, not just a position.

Explore the arguments Substack

Interactive argument atlas

Three essays, viewed as dynamic systems.

Change the essay to see its core thought experiment. The same visual language—agents, information, and time—reveals where their arguments connect.

Read the summaries

Education / coordination / search

Independent explorers become correlated around the same legible path, leaving more of the search space untouched.

Tip: move across the diagram to perturb the system and compare local choices with the global pattern.

Featured essays

The argument behind each title

Each summary exposes the thesis, mechanism, and unresolved question before you commit to the full essay.

01Education / game theory

Tragedy of the Correlated Commons

When individually sensible choices collapse a society’s search strategy.

Prestige-oriented educational choices can be locally rational while making everyone explore the same narrow region. The loss is not only diversity as an aesthetic good; it is reduced coverage of the possibility space from which new expertise, institutions, and discoveries emerge.

Thesis

A population can over-optimize for legible, low-variance outcomes and underinvest in unusual depth-first paths.

Mechanism

Shared status signals correlate family strategies. Correlation crowds known peaks and removes the portfolio benefit of independent search.

Open question

Which institutions can protect exploration without asking individuals to absorb all of its risk?

Prestige as a coordination signal
Correlated search
The frontier cost
Restoring exploration

02Economics / optimization

The Communist Constraint

Central planning as a computational bottleneck.

The essay reframes the coordination problem in optimization language. When local conditions are high-dimensional and continuously changing, the difficulty is not simply finding well-intentioned planners; it is collecting fresh information, computing an update, and applying it before the world changes again.

Thesis

Some economic limits are best understood as constraints on information bandwidth, latency, and objective specification.

Mechanism

Distributed actors produce noisy but frequent local updates. A central optimizer receives a cleaner-looking batch only after delay and compression.

Open question

Where can modern computation improve coordination, and where does it merely move the bottleneck?

The optimization analogy
Information and dimensionality
Stale gradients
Markets as distributed updates

03Genetics / long horizon

Compounding Returns of Genetic Capital

Small changes become different questions when they persist across generations.

Many technologies are evaluated through their immediate effect. Germline interventions force a longer accounting: a small directional change can propagate, interact with assortative choice and institutions, and alter the starting point of subsequent generations.

Thesis

Intergenerational persistence makes the total effect of selection qualitatively different from a one-time intervention.

Mechanism

Heritable changes alter the base on which later choices act; even modest increments can accumulate over a long horizon.

Open question

How should governance weigh distant compounding, uneven access, uncertainty, and individual autonomy?

From capital to inheritance
The compounding model
Technology and selection
Governance under long horizons

Common grammar

Optimization is a lens, not an answer

Across the essays, optimization language makes hidden trade-offs easier to name: exploration versus exploitation, fresh local gradients versus delayed global plans, and immediate effects versus compounding state.

Search spacesInformation latencyCoordination failuresLong horizons