Build One Good Model

Planet-scale problems create planet-scale language.

Global systems. Global restoration. Global transformation.

The ambition may be right. The starting point can still be wrong.

Trying to solve everything at once makes it easy to build abstractions that look impressive from a distance and become useless when they meet a real place.

I am increasingly drawn to the opposite approach.

Start somewhere specific

Choose one person, one place, one value chain or one problem.

Understand it deeply. Find the strongest existing knowledge. Find the people already doing the work. Find what is missing.

Then build the smallest useful intervention you can test in reality.

Measure what happened. Correct what was wrong. Only then ask what should repeat.

Reusable does not mean identical

A useful model has two layers.

One part should become easier every time: how you organise evidence, show uncertainty, structure a project, measure an outcome or connect people.

The other part must stay local: species, culture, politics, geography, economics and people.

Standardisation should make adaptation easier, not erase difference.

Build with the world that already exists

If the best answer already exists, use it.

If a field group has spent twenty years learning a landscape, a new platform should not arrive pretending to know more because its interface is cleaner. If a public dataset is authoritative, connect to it. If a company already solved a technical problem, help that solution reach the places where it is useful.

Build the missing connective tissue.

One excellent shop before ten thousand

The analogy I keep returning to is simple: build one excellent shop before trying to build ten thousand.

Not because ten thousand is too ambitious.

Because the first one teaches you what the plan did not know.

Reality is where architecture becomes knowledge.