← All notes
Owning your data

A twin you can only open in their tab is a rental

A digital twin is a working likeness of a place or a process, kept in sync with the thing it represents. The definition does not say the likeness has to live on a vendor's server. If the model disappears when the login does, it was never yours.

Escherbridge

Most products that call themselves a digital twin start with a quiet assumption. The model of the place — the terrain, the equipment, the sensor feeds, the decisions — will live on their servers. You get a browser tab, a login, and a bill. If the service is reachable, the likeness is there. If it is not, the likeness is gone, even if the place itself is fine.

That is a convenient way to run a demo. It is a weak way to hold a working model of something you actually have to operate.

The phrase "digital twin" is used loosely, so it helps to start with a definition that is not a marketing page. In a 2023 Winter Simulation Conference panel paper, researchers including Guodong Shao of the National Institute of Standards and Technology restated the manufacturing definition from ISO 23247: a digital twin is a "fit for purpose digital representation of an observable manufacturing element with synchronization between the element and its digital representation." The point of the sentence is not the brand. It is the two jobs. There is a representation, good enough for a purpose. And there is a way to keep that representation in agreement with the thing it stands for.

Notice what the definition does not say. It does not say the representation has to live in a vendor's datacenter. It does not say the authoritative copy is the one behind someone else's login. Synchronization is a requirement. Centralization is a design choice, and often an unexamined one.

A separate argument, older than the current wave of twin platforms, is about where the primary copy of work should sit. In 2019, Martin Kleppmann, Adam Wiggins, Peter van Hardenberg, and Mark McGranaghan set out a case for "local-first software": software that keeps collaboration without making a remote server the home of the work. Their opening is still the clearest version of the problem. Cloud apps make it easy to work from several devices. "However, by centralizing data storage on servers, cloud apps also take away ownership and agency from users. If a service shuts down, the software stops functioning, and data created with that software is lost."

The software world calls this 'local-first': the copy on your own computers is the real one, and the vendor's servers are a helpful spare. Servers can still exist, as secondary copies, to help with access from more than one machine. Among their ideals are that the network is optional, that the work should survive in the long term, and that the people who made it "retain ultimate ownership and control." They were writing about documents, drawings, and other files a person makes. The same prior applies to a likeness of a place. If the twin is a working model, it should still be a working model on a day when the wide-area link is down, or when the vendor has changed its terms, or when the contract has ended and the export wizard is a support ticket.

Local-first is not nostalgia for a file on a laptop that nobody else can see. The 2019 essay is explicit that collaboration is part of the goal, not a feature to be given up in exchange for ownership. The question is which copy is in charge. In a cloud twin, an edit that does not reach the server "didn't happen." In a local-first design, the edit happens on the machine in front of you, and synchronization follows when it can. For a model of a site, that difference is not philosophical. Sites have bad networks. They have shifts that do not wait for a spinner. They have people who need the current geometry, not a cached scene that might be stale and might be gone.

The 2023 panel paper makes a second, quieter point that is easy to miss in a product launch. A twin has a lifecycle of its own, "starting from specifying the requirements of the digital twins until their decommissioning." The representation has to be maintained, or it stops being fit for purpose. A rental tab makes that maintenance look like someone else's problem, until the day the likeness and the place have drifted apart and nobody can say which one to trust. Ownership is not only the right to copy a file. It is the duty to notice when the file is no longer the thing.

Remote compute still has a job, and so do backups, and so does a second copy in a safe place. Kleppmann and colleagues called servers "cloud peers": a supporting role, not the source of truth. The point is narrower. If the only way to see the twin is to ask a vendor whether you may, then you do not have a twin. You have a visualization with a landlord.

This is a different problem from a file that leaves one tool and arrives incomplete in another. Interchange can fail even when you do hold the file. The problem here is prior to that: whether you hold the file at all, and whether the software that makes it useful still runs when the service does not.

Fractal Engine, the organizational-memory layer in the PrimusNeo stack, is being built with that prior in mind. It is alpha. There is no versioned release. Its public README describes a 3D editor, models stored as glTF with BLAKE3 hashes, object history, peer-to-peer sync, and a plugin host that starts but does not auto-load packages. A hash will tell you whether a file changed. History will tell you that a change was recorded. Neither, by itself, makes a complete digital twin. The layer is not claimed to be running at a particular site. What it can offer is smaller: a place to keep a model and its memory on terms that do not begin with someone else's cloud.

A twin should be a likeness you can hold. Useful because it stays in sync with the thing it represents. Durable because the people who need it can still open it when the network is down, or the vendor is. Shareable because copies can travel without the original having to leave home. If it only exists in their tab, it was never yours. It was a rental with a nicer name.

Sources

  1. Alp Akcay, Stephan Biller, Boon Ping Gan, Christoph Laroque, and Guodong Shao, “Maintenance and Operations of Manufacturing Digital Twins,” Proceedings of the 2023 Winter Simulation Conference
  2. Martin Kleppmann, Adam Wiggins, Peter van Hardenberg, and Mark McGranaghan, “Local-first software: you own your data, in spite of the cloud,” Ink & Switch / Onward! 2019
  3. Escherbridge/fractalengine README

Related