Escape
In Edwin Abbott’s 1884 story Flatland, a sphere passes through a two-dimensional world. The people who live there see a circle appear, grow, shrink, and vanish. The sphere was whole the entire time. They could only see the slice of it that crossed their plane.
We live the same way. Not in two dimensions, but inside systems we only ever see a slice of.
The goal
Make the world’s systems visible, so we see the consequences before we cause them.
We want a world where people can see how the systems they live in work, play forward the odds of what happens if those systems change, and argue about assumptions instead of conclusions. And we want that to hold even when machines build the models.
Why it matters
A business, a supply chain, a household budget, a carbon plan, the curve of an epidemic: each is a system. Each has inputs, transformations, and outputs. Each has a few variables that decide a lot and many that decide little, effects that arrive long after their causes, and a range of possible outcomes rather than one. Sometimes a system is measured in money. Often it isn’t.
We act inside these systems every day, but we see them one slice at a time: this month’s number, last quarter’s chart, one row of a sheet. The consequences arrive anyway, usually after we’ve caused them.
Seeing changes that. Weather forecasts have gained about a day of useful lead time every decade since the 1980s.1 In Bangladesh, cyclones killed an estimated 300,000 to 500,000 people in 1970. With forecasts, shelters, and evacuation working together, Cyclone Sidr killed about 3,400 in 2007, and Cyclone Amphan about 26 in 2020.2 No one changed the weather. People changed what they did, because they could see it coming.
The systems we build differ from the weather in one way that matters: we can change them. A plan can be redrawn, a price reset, a policy withdrawn before it does harm. But only if someone sees the consequence first.
Why now
The world’s systems were already hard to see. They’re about to get much harder.
Agents are starting to do the work. They’ll build the models, run the numbers, and act on the results. Every task they take on makes the systems faster, more detailed, and more connected. Complexity goes up, and what any one person can see goes down.
That’s the fork. Down one road, people hand over the picture along with the work, and understand less of the systems they live in every year. Down the other, they have an instrument that keeps the system in view, however much of the work the machines do.
We’re building for the second road. This is for people, not agents. Agents are welcome to build the model. The point is that a person can see it, step inside it, and judge it.
What stands in the way
Most of the world’s quantitative reasoning still lives on a flat grid. A cell that says 15% doesn’t know what it means, what depends on it, or what happens if it’s wrong. The structure exists, but only in the head of whoever built the sheet.
Agents make this worse before they make it better. When writing a model costs nothing, the cost moves to checking it. A flat sheet an agent wrote in thirty seconds can take a person an afternoon to trust.
Our bet is that the form cheapest to check will win. A model whose structure, course over time, and possible futures are visible should be far easier to trust than one that has to be reverse-engineered cell by cell. If that’s true, no one has to be persuaded to see their systems. The world will move toward visible models on its own, because they’re cheaper to trust.
That’s a belief, not yet a fact. The plan starts by testing it.
What we’re building
A model that leaves the flat grid behind: one that knows what each number is, what feeds it, how it moves through time, and how uncertain it is. It follows the laws systems obey, starting with the basic ones: units that are checked, stocks and flows, delays, and uncertainty. Agents can write it. People judge it.
And an instrument to see it. Inside one, you see a system’s structure, its course over time, and its possible futures. You change an assumption and watch the consequence travel. You find the vital few among the trivial many: the handful of things that decide the outcome, apart from everything that only looks urgent.
No one can understand everything, everywhere, all at once. But anyone should be able to see what decides the outcome. What they get isn’t a prophecy. It’s the odds, with the levers attached.
The plan
- 01
See inside one model.
Build an instrument that lets one person see inside one model of any system. This is flatland desk.
- 02
Prove that seeing beats a flat sheet.
Test it in public, with the rules registered before each test runs and the results published whatever they show. If the flat sheet wins, we’ll say so, and this plan will change.
- 03
Make visible models travel.
Publish the model format, so any agent can write a model people can see, and let every model be handed to someone else as something they can open and play.
- 04
Hold models to reality.
When reality arrives, show where it landed among the futures the model drew, so models earn trust by being measured.
- 05
See bigger systems.
Let models compose into larger ones, and cover more of the laws systems obey.
The instrument pays for the road. What people pay for in flatland desk funds the proof, the format, the track record, and the research. If we’re right, every visible model creates more people who want an instrument to see one, and the loop turns.
As this is written, we’re at the start: phase one is in public alpha, and the first test of phase two is running.
The escape
In Abbott’s story, a square is lifted out of his plane and sees his world whole for the first time. That’s the escape we mean. Not out of the systems we live in, but far enough above them to see how they work, while there’s still time to change them.
We’d rather not make it alone. Try flatland desk on a system you care about, take the test, and tell us what you see: info@flatlandfi.com.
The labs are giving machines a model of the world. flatland gives people a view of it.
- Bauer, Thorpe, and Brunet, “The quiet revolution of numerical weather prediction,” Nature 525, 47–55 (2015). ↩
- Sammonds et al., British Academy: thebritishacademy.ac.uk. 1970 was in what was then East Pakistan. ↩