Own the rails. Then own a piece of what runs on them.

Build the infrastructure, not just the businesses.

The group runs an agent mesh, an identity layer, a settlement ledger and an open standard for machine-readable organisations. Its own companies are the first customers, which is the only honest way to build infrastructure: the failure modes arrive in your own operation before they arrive in somebody else’s.

Every position the office underwrites is tested twice — once as an investment, and once as something the estate might otherwise have to pay a third party for. A holding that becomes a dependency pays back on both.

Permanent capital is what makes move one possible.

Infrastructure pays back over fifteen years. A ten-year fund cannot hold it — it must sell its best asset in year nine whether or not year nine is the right year, and everyone in the market knows the date it has to sell by.

This office has no fund life, no vintage year and no date on which anyone must be repaid. It is never a forced seller, and it can decline an exit that arrives at the wrong time. The horizon that follows is generational rather than decadal — longer than any instrument that could be sold to this office — which is what allows it to underwrite a payback nobody with a fund can wait for. That is not a preference; it is the structural precondition for owning rails at all.

The next readers of a company are machines. Almost nothing is written for them.

Every organisation in this estate publishes a machine-readable charter declaring which functions its software agents perform, what each may do, and the threshold above which a human must approve. It publishes a handshake saying who it is and who answers for it. Both are open, checkable, and served at a fixed path.

The specification is Apache-licensed and given away, because a standard only its author runs is a product. The group wrote it, applied it to itself first — the parent was the last entity in the estate to conform, and that was the first finding of its own audit — and gated it in CI so it cannot quietly stop being true.

A capability built once should be inherited by everything the office controls.

This is the move a conventional holding company cannot make, and it is the reason the first three are worth making. A group that owns companies gets two things from the parent: capital allocated well, and managers left alone. A group that owns companies and the layer they run on gets a third — an improvement built once arrives everywhere at the same time.

When the identity layer gets better at proving whose authority an agent carries, every company in the estate can prove it, on the day it ships, without any of them writing anything. When the charter standard gains a way to declare a new function, every organisation here can declare it. A payment path, a procurement flow, an approval gate, a way of describing a role — each is built once, in one place, by whoever needed it first, and inherited by the rest.

The condition is ownership, and it is why the percentage matters more than the count. A minority position cannot be told to adopt anything; a controlled company can inherit by default. That is the argument for owning most of a smaller number of companies rather than a little of many — not concentration for its own sake, but because inheritance only works downhill.

Nothing here is a forecast about how large that becomes. It is a mechanism, and the mechanism is either running or it is not: the test is whether a capability built in one company is in production at another without that company having built it. The mechanism has a name and an institutional definition — the propagation premium — and the six ways this office takes a position, including the one test for giving ownership up, are set out under ownership.