twins.sgit.ai — an interface to reality, not a simulation of it
The industry's digital twin is a simulation of a physical asset, valued for how faithfully it copies. This one is different, and the difference is the point: a twin here is an interface to a real thing, valued for whether it is connected. It is where a graph stops modelling and starts touching the world.
Source: https://twins.sgit.ai/index.html↗ · site v0.1.0 · markdown twin of the front page.
The definition
"a digital twin is, in essence, a system that has properties, behaviours, functions, and inputs and outputs, and we can define all of those."
And its generality, which is the sentence worth memorising: one can be made "of anything: an organisation, an element, a mail system, an inbox, a person, a behaviour, an event, an action, external factors like weather, even luck." Non-determinism is included rather than excluded: "the fact that in the real world they are not deterministic, or maybe random, we can capture that."
Why the graph needs one at every edge
These graphs refuse properties: "properties do not have meaning, they are just words; we capture meaning through connectivity." Which creates a problem at the edge. If meaning is connectivity, where does connectivity stop?
"the power of the twin is that we always arrive at the twin, so the edges and the peaks and the endpoints of the graph continue into the twin, and then ideally into reality."
Every terminal node is a twin, and the twin is "the doorway from the model to the real
system." This is why the grounding ladder reads `Measure := an observation of the node it
measures, grounded on a Twin`, and why the established edge grammar ends with "the
connection from any node to a twin and onward to reality is connected_to."
Four ideas that are original here
- Connectedness is a measurable fact ↗ — "whether we can continue to reality is a measurable fact, it is connected or it is not." Where it is not, the state is recorded as a tracked air gap. Applied to a legal instrument, that becomes a computable coverage measure.
- The discipline of reality ↗ — "everything has to be relevant, everything has to be a fact, everything has to exist ... either we have evidence and it exists, or we do not." No speculative risks polluting the graph.
- Twins as actors ↗ — "every action that updates the graph, including accepting a risk, is performed by a twin", with reasoning documented and a persona carried.
- Twins make risk testable ↗ — "the same techniques we use with software", turned on an organisation's risk, because the twin gives them something to run against.
The build state, unsoftened
| Artefact | State | What it proves |
|---|---|---|
| The S3-compatible vault container | Working | A twin is a drop-in interface: "code using boto3 believes it is talking to real S3", and the consuming code does not change |
| The AWS twins in the IAM config risk engine | Partial | Facts from twins, judgement elsewhere. Python, unit-tested, JSON out |
Twin in the grounding ladder, connected_to in the edge grammar | Established | Cited across the estate |
| The 2FA demo, the world model, the org twin, the agent twin | Designed, unbuilt | Nothing yet, and labelled as design everywhere |
One working twin, one set of half-built twins, two established primitives, and a large designed layer. The full table ↗.
The one that works
"the container is a digital twin of S3, presenting the same S3-compatible API, so that code using boto3, the AWS CLI, or any S3 SDK believes it is talking to real S3, while the files are served from a vault, from local disk, or from memory, chosen as a swappable backend; the service's own code does not change."
Its value is not that it simulates S3 well. It is that boto3 cannot tell the difference. The worked proof ↗.
The naming collision, ruled on
A brief arrived in August calling a credential-holding execution broker a "Service Twin", and the corpus's own review said: "the name collides, because twin already means something specific in this corpus." It does, and in the dimension that matters most: a corpus twin holds no credentials, and the broker's whole function is to hold them.
The ruling: twin keeps its corpus meaning, and the broker is presented by its function. The collision itself is published, and enforced by a name-watch in the release gate. The ruling ↗ · the broker ↗.
Published unresolved
Five open questions travel with this material. The two that matter most: how connected a twin really is once staleness is admitted (binary connectedness is what makes coverage computable, and a twin updated "manually once a week" is connected and six days stale), and whether the broker's concentration risk is acceptable, since it "inverts the catastrophic failure property the rest of the architecture depends on." Gaps and open questions ↗.
Who is writing this
Published by the sgit project, which builds the graph model and the vault layer the twin primitive sits inside. A participant publishing a definition, stated upfront. Where this approach loses ↗ · what this site owns and what it does not ↗.
All content CC BY 4.0. The brief pack this site is built from is published verbatim at /briefs/ ↗. Machine-readable summaries: /llms.txt↗ and /llms-full.txt↗.