A digital twin is a mirror. It models the network's state — topology, KPIs, flows, the lifecycle of every cell and subscriber. Useful, but mute. It shows you a thousand numbers and waits for a human to work out which ones mean something, and what to do about it.
A Reality Twin goes one step further. It reads that mirror, pulls in context from the operational apps around it, and writes the situation down — a reasoned account, in words, of what is actually happening and why.
It narrates the situation
Not a dashboard of metrics — an explanation. The Reality Twin takes the raw state and turns it into the kind of paragraph an experienced operator would write at 2 a.m.:
“Cell 4521's S1 link broke at 14:02. Downlink drop-rate has climbed to 12% and is still rising. VoLTE setup delays on the enterprise APN now exceed 800 ms; retries are amplifying the load. The platinum-SLA segment breaches in roughly 18 minutes unless traffic steers to neighbor 4523.”
That is a Reality Twin narration. Some of the situations it writes up:
- Link break / cell down — the fault, and what is inheriting it downstream.
- High drop-call rate — where, how fast it's climbing, who it touches.
- Latency & delay cascade — one slow hop rippling into everything behind it.
- SLA breach forecast — which segment, how long until it breaks, what prevents it.
- Congestion — PRB saturation collapsing throughput for a named group of subs.
- Coverage hole — a site outage no neighbor can fill, over the place it matters.
- Interference / PIM — a rising noise floor that is not load and won't fix itself.
- Backhaul degradation — a flapping transport hop and everything that sits behind it.
- Energy anomaly — an off-pattern power draw that reads as fault, not traffic.
- Churn risk — high-value subscribers hit by the same quality event (from CRM).
- Security correlation — an auth-failure burst tied to a config change minutes earlier (from SOC).
- Provisioning storm — activation failures climbing after a release, subs stuck pending.
Where the narration comes from
The Reality Twin doesn't invent the story. It assembles it from two things:
- The digital twins — the simulators (coverage, CRM, demand, incidents, telemetry, MCX, core provisioning…). They mirror the network's real state and dynamics. They are the ground truth the narration stands on.
- The context builders — the operational apps (NOC, SOC, RAN/MAP, IOC, CRM, growth…). Each is a contribution surface: it enriches the situation with incident-shaped, security-shaped, customer-shaped, RF-shaped context. The constellation is a context-building factory; the narration is what it adds up to.
Digital twin says what the state is. Context builders say what surrounds it. The Reality Twin says what it means.
A sub-world of possible worlds
A digital twin can be in many states. From it you can reason into any number of possible worlds — what if this site drops, what if that campaign lands, what if traffic doubles. The Reality Twin is the one that gets narrated as real: the sub-world that is actually the case right now, picked out of the space of possibilities and written down. (The same machinery narrates what-ifs too — that's how you rehearse a response before the situation is real.)
Then the loop closes — the self-driving part
The narration plus its prepared context is the thing agents read. They don't parse raw telemetry; they read the situation as written, reason over it, and act — steer traffic, reconfigure, open or close a ticket, escalate. The action changes the network, the digital twin re-mirrors it, the Reality Twin re-narrates, and the agents read the next account.
That loop is the agentic, autonomous, self-driving approach. The Reality Twin is not the loop — it is what makes the loop legible enough to be autonomous. Runtime + autoresearch tunes the agents; the factory is the library they come from; DOIL is the language the whole thing is written in.
Why it matters
A digital twin asks “is the model accurate?” A Reality Twin asks “do we understand the situation well enough to act without a human in the loop?” Those are different questions — and the second is the one a self-driving NOC has to answer.
Practically, it shifts the effort. Less on perfect mirror fidelity, more on the reasoning that turns state into a trustworthy account. Less on dashboards a human stares at, more on narrations an agent can act on. The mirror was never the hard part. The hard part is reading it and saying, in words, what to do.
Where to go next
If this frame works for you, the rest of the site is the receipts:
- The simulators — the digital twins the narration stands on.
- The simulator cascade — how the twins feed each other into a fuller picture.
- Synthetic flows — authoring a situation in DOIL.
- Composite indexes in DOIL — how the cockpit reads state.
- Autoresearch — how the acting agents get tuned.
- Introducing DOIL — the language itself.
Digital twin: the mirror. Reality Twin: the account of what the mirror shows. The loop around it: the self-driving NOC.