Chapter 29. Rehearsing the future

A working draft, November 2026. This book is being written while The Grove is being built; expect it to change. Checked against the build as it stood on 1 November 2026.

In this chapter

• Why pilots, consultants' maps and digital twins each leave a gap before a change is lived
• How organisations are grown by simulation, with simulated people in a simulated world, over days
• What a rehearsal is for: finding where structure creates friction or freedom, not picking winners
• Four decisions that came from watching simulated organisations go wrong
• What simulation can't tell us, and what we've decided but not built

The sceptic's objection here is short, and it's a good one.

You can't test an organisation before living it. An organisation is people, and people aren't predictable. Simulations of people are toys. They tell you what you put in, dressed up as discovery. Whatever a model says about how your team will behave, your team will do something else on Tuesday.

We agree with most of that. A simulation won't tell you what your organisation will do. Nothing will. But there's a narrower question it can help with, and it's the one that costs most when it's answered too late: where will this structure rub? Where will work sit with nobody holding it, where will a rule that looked fair on paper hand one person too much, where will people have to work around the system instead of through it? Those are questions about the structure, not about the people, and a structure can be rehearsed.

What finding out the hard way costs

Most organisations find out how a change works by making it.

A new way of deciding is announced at an away day. A team is split in two. Then everyone lives with it for six months, and the cost of the parts that didn't fit is paid by the people inside: the decisions that stalled, the work that fell between the new teams, the quiet person who ended up doing the admin for both. By the time the friction is visible, it has become "how things are here", and changing it again feels like failure.

There are three common ways to try to know sooner, and each leaves a gap.

A pilot runs the change for real, in one team. It's honest, because real people live it. But it's slow and costly, it can only be run once, and it tells you one future. If the pilot team happens to be unusually generous, or unusually tired, you learn about them as much as about the change. Nobody can run it again with the same people and one thing different.

A consultant's map draws the organisation as it is: boxes, flows, a diagnosis. It can be insightful. But a map is still. It shows where things are, not how they'll move once the change starts pushing on them.

A digital twin, borrowed from engineering, mirrors a live system closely enough to predict and optimise it. That works for a jet engine. For an organisation of people it promises something nobody can deliver: a faithful copy of how humans will choose.

Growing an organisation instead of drawing one

There's an older tradition that asks a humbler question. In the early 1970s Thomas Schelling showed, with nothing more than counters on a grid, that a town of people who each mildly prefer a few neighbours like themselves can sort itself into sharp divides nobody wanted.¹ The lesson wasn't a prediction about any real town. It was that small rules, followed by many people, make patterns none of them chose. Joshua Epstein and Robert Axtell took the idea further, growing whole artificial societies from simple agents on a landscape, to see which structures emerge from which rules.²

That is the spirit we work in. Alongside the system itself we keep a rig we call the harness. It grows an organisation by simulation: simulated people, in a simulated world, over days and sometimes months, using the system the way anyone would.

The people come from a pool of would-be members. Each has a temperament somewhere on a range, from generously supportive, through steady and transactional, to self-serving and, at the far end, predatory. Each has hours to give, things they're good at, and people they know, mostly people like themselves. Nobody joins by magic. They come only when someone already inside invites them, so who gets invited shapes who the organisation becomes. A self-serving person is planted close to the founder on purpose, because real organisations have them.

Some runs have a language model play each person, agenda and flaws included. Others play them by simple rules, which costs nothing and replays exactly. Either way, the simulated people can do only what the page offers a real person, and see only what a real person in their place could see. The harness has no back door into the system. If a simulated person does something no person could, that's counted as a fault in the harness, not a finding about the organisation.

Worlds with their own truth

An organisation needs something to be about, so the harness gives it a world.

The first was a small family kitchen: a young man selling his grandmother's baking, first at a car boot sale, then on a market stall, perhaps one day from a second kitchen. The second was a practice of facilitators who run group sessions for other organisations and one-to-ones booked by email. Others followed: a coaching collective working with charities, a small design studio, and a cooperative spread across four cities, branches four deep and a lot arriving every day.

Each world keeps its own truth. In the kitchen it knows what was really baked and sold, how good it really was, where the cash really went. The organisation knows only what its people chose to write down. The gap between the two is one of the things we watch, because that gap is where value leaks in real organisations too.

The world also writes in. Customers order and complain. Suppliers raise their prices. A market organiser offers a pitch and wants two people on the stall first. An inspector asks what's written down about allergens. In the facilitators' world, a client's HR director asks who came to the sessions: saying no holds the line on confidentiality, saying yes crosses it, and saying nothing loses the client. None of this tells anyone what to do. The organisation answers as it answers anything, and the world reads the answer the next morning.

We keep the world as written rules, not as a model telling a story each day. A storyteller would invent its own truth and couldn't be replayed.

Freezing a moment, and a fresh night each night

Two things make this a rehearsal rather than a performance.

The first is that a moment can be frozen. A founder's first hour with Wise-Eye, planting the organisation and bringing their work in, can be stopped and kept exactly as it stood. From there it can be grown on many times, each with a different run of chance, so one founding can be seen weeks later in a dozen futures. A thin start that looked fine in the hour shows itself in week six. The moments we grow for people trying The Grove for the first time work the same way: each is written down as its world, its starting chance and its day, and grown afresh wherever it runs. The same start grows the same organisation, day for day, on any machine.

That's also how most people will first meet The Grove. Before it goes live, the way in is a sandpit: a simulated organisation, grown up to a key moment in its life, where you step in and take the part of one of its people. A proposal is about to be decided, or work is arriving faster than a branch can hold, or a founder is about to share the root, and you're in the room for it. You can try things, get them wrong and start again. Nothing you do there touches a real organisation. It's a place to play, and to feel how it works from the inside, before anyone has to stake anything on it.

The second is that every night the harness runs on fresh ground. A new starting chance each night means a new stretch of possibilities: the kitchen and the facilitators' practice, a flood of mail and people who say yes to everything, a flood of ideas from below, a business that already runs laid in on its first day, a handful of first hours. Alongside them, drills walk each area of the rules step by step and check after every step that nothing has come apart. No model is called on those nights. Every run replays exactly from its starting chance, so anything strange can be seen again before anyone acts on it.

What the night looks for is friction. Most of what a run says is about the simulated society, not the system: that a self-serving person ended up holding the root, that nobody took up a measure. Those are read as questions about that society, never as a score. What matters is narrower: the system refusing something it offered, the books not adding up, structure running ahead of the work, the same work born twice, work left with no home.

Friction, not winners

This is the heart of it. A rehearsal doesn't ask which organisation wins. It asks where the structure gets in people's way, and where it gives them room.

One habit captures it. When a promise to someone outside is missed in a run, the harness puts it down to one cause: the harness itself, a rule of the simulation, the world, the people, or the system. Shown to someone who had a turn and not done is the people's. Shown to nobody who could act is the system's own, and only that one is ours to fix. It keeps us honest about what a simulation can and can't blame on design.

Martin Pfiffner, who has spent years putting Stafford Beer's viable system model to work in real companies, treats an organisation's structure as something to work through deliberately before people have to live inside it, rather than something inherited and patched.³ That's what the harness lets us do with the rules themselves. And George Box's warning, that every model is wrong and some are useful, is the right one to keep close.⁴ We're not trying to be right about people. We're trying to be useful about structure.

What it has changed

Some of the system's shape came straight from watching simulated organisations go wrong.

A promise made to a customer landed on a branch where nobody held it and its date drove nothing, so a promise is now carried by a named person, on a clock, and rises when it slips. An organisation under pressure was offered a split, then offered back together, then split again, so a job that has just moved now settles for eight weeks before anything offers to move it again. Budget allocations passed as quiet grabs because they carried only an amount, so a budget proposal now carries its evidence: the branch's pressure to grow, its load of work and what the money is for. And Wise-Eye, once trusted, confirmed orders on its own that a person would have paused over, so it may now send a promise by itself only when it sits within what's usual for that kind of work; anything outside waits for a person, with the reason shown.

The same runs are where the organisation's learned sense of itself, in Chapter 21, is first being taught, and the same caution holds: rehearse, don't foretell. An instinct learned only from simulated people is a model of a model, which is why it's checked against people who choose for themselves before anything is trusted.

What a simulation can't tell you

The limits are real, and we'd rather list them than have you find them.

Simulated people are models of people. A cheap model plays nice; we give the self-serving characters a stronger one because otherwise everyone is agreeable. Rules-played people are blunter still. Neither feels the sting of having a task taken from them, or the relief of a seat finally shared.

There's a deeper worry, and we've asked it of ourselves: if the system is tested against the harness, and shaped by what the harness shows, can the testing be trusted? Part of our answer is more than one world, so that what was fitted to one kitchen shows up as not fitting a facilitators' practice. Part is a hard line that the worlds live in the harness and never in the product. Nothing from any world appears in the system's code, and the guidance Wise-Eye reads carries principles, never cases; a check refuses a world's details where the model reads. The system's own tests never call a real model, and neither do the nightly runs. Model-played runs are separate, paid for separately, stopped at a hard budget, and the key never enters the record.

And some things can't be learned without real organisations at all: whether a forum trusts the readings, how a clock on a promise feels on a bad week, whether people will grant Wise-Eye standing and take it back. A simulation can show where a structure rubs. Only people can show whether the rubbing matters.

What we're building towards

The obvious next step is to turn this towards a real organisation. Before a big change, take a frozen copy of a live organisation, let the proposal play out on the copy many times, and see where it creates friction or freedom. Not a prediction of what will happen, which no living thing allows, but a rehearsal.

We've decided the direction and parked it. On 1 November 2026, freezing a live organisation to rehearse a proposal was set aside until the core has settled with real people in it. When it comes, the intent is that it belongs to the organisation's own forum: something people deciding a change can ask for, read and argue with, never a verdict handed down from outside.

The groundwork exists as an experiment. The harness can already take an organisation at a given day and play many futures from it, with each possible change given its own futures on the same runs of chance, so the difference you see is the change and not luck. Those forecasts are sealed before they're scored against what happened, and they're shown to no organisation until forecasts like them have been measured against reality. The only kind we'd ever point at a real organisation is the one that knows people as a stranger would, from what they do, never from a temperament someone wrote down.

How we decided

We weighed three ways of knowing sooner. Pilots alone were too slow, couldn't be run twice, and put the cost on real people first. A model narrating each day was cheap to build and impossible to trust. A faithful twin of human behaviour promised too much.

We chose to grow organisations through the system's own front door, in worlds with a truth of their own, replayable from a starting chance, and read for friction rather than outcomes. We chose more than one world, because one world can't tell what generalises from what was fitted. And we chose to keep every world out of the product, so the system can't quietly learn to pass its own test.

What's uncertain is how far the rehearsal will travel. Simulated friction has pointed us to real flaws in the rules, each traced by hand before anything changed, but none of it has yet been checked against a real organisation. Whether a rehearsal on a frozen copy of a living organisation helps its forum decide better, or just gives an argument a chart to hide behind, is something only use will show.

What you might take from this

You can't rehearse people. You can rehearse a structure, many times, before anyone pays for it, and look for where it rubs rather than who wins.

A rehearsal isn't a forecast: it finds the friction before your people have to.

Notes and further reading

On Stafford Beer's viable system model, which Pfiffner builds on, see Chapter 20. For simulation as a way of asking "what structure produces this?" rather than "what will happen?", Epstein's later essays, collected as Generative Social Science (Princeton University Press, 2006), are a clear and readable place to start.