Real systems, one at a time.
The Systems Atlas Project maps individual systems as they actually operate — a particular court, for example: who is in it, how it works, and what it produces — and compares them with the same kind of system elsewhere. The aim is to find changes that would make a system work better, and to say how likely that is on the evidence.
Starting again Nothing published yet
How the work runs
Map it, compare it, then test a change.
-
Not started
Map
One real system as it operates: who is in it, what each part does, and what it produces. Every claim sourced.
-
Not started
Compare
The same kind of system in other places, on measures stated in advance, so that a difference is a finding rather than an impression.
-
Not started
Assess a change
A candidate change, the measure it is meant to move, the evidence for it and how strong that is, the conditions it depends on, and what else it touches.
What "better" means here
Against a named measure, with the uncertainty attached.
Not an ideal system. A change counts as better only against a measure named in advance — the time it takes to reach a hearing, say, or how often a decision is overturned — and the claim is an estimate with its uncertainty, not a verdict.
A change measured in one place does not automatically carry to another. The map of each system is what shows whether the conditions the change relied on are present there too.
Where nobody knows what a change does to something else, that is stated as a gap rather than left out.
Where it came from
The first approach is archived, not deleted.
The project began as an attempt to decompose every system that shapes the world into its parts, from the top down. That approach was set aside on 27 September 2026, as too ambitious to finish and aimed slightly wrong. Everything it produced is kept, along with why it was set aside.
Contributing
Open to being told it is wrong.
This is a project led by one person, and it is open to being argued with. Until the new work has something to argue with, email is the way in.