A device that goes quiet is a signal.
Telecom and IoT have the most devices and the most irregular reporting: some report every second, some once a day, some stop. AnantState tracks each device individually and uses the actual time between its messages, so silence is noticed.
In plain termsIt treats a device that stops talking as news.
Silence, seen across a gateway
An illustration of three devices going quiet behind one gateway.
In most monitoring a device that has stopped reporting looks the same as one reporting healthy values. In a state model, silence is a divergence the model can see, and silence that clusters behind one gateway points at the gateway.
Devices that go quiet together · illustrative
- Gateway 17linked to Meter 1 · reporting, Meter 2 · reporting, Meter 3 · quiet, Meter 4 · quiet, Meter 5 · quiet
- Meter 1 · reportinglinked to Gateway 17
- Meter 2 · reportinglinked to Gateway 17
- Meter 3 · quietlinked to Gateway 17
- Meter 4 · quietlinked to Gateway 17
- Meter 5 · quietlinked to Gateway 17
Time matters more here than anywhere
Devices report on their own clocks and go silent for reasons that are sometimes meaningful and sometimes a low battery. The engine advances state by the observed time between events. When a gap exceeds the ceiling it was trained on it falls back to a documented default, because extrapolating across a silent device produces a confident number with no information in it.
What moves across a fleet
| State | Baseline drivers | Unexpected drivers |
|---|---|---|
| Connectivity health | Coverage, load, scheduled maintenance | Interference, backhaul degradation, misconfiguration |
| Usage behavior | Plan, season, device type | SIM misuse, device compromise, subscription fraud |
| Device lifecycle | Age, firmware, environment | Battery degradation, physical damage |
| Fleet integrity | Provisioning pipeline | Unmanaged devices, credential sharing |
At fleet scale
- Two million devices and forty thousand alerts a day make per-device alerting useless without an ordering by value at stake.
- Provisioning anomalies become security incidents later, and the lag between an unusual pattern and its consequence is what a forward state model is for.
- Missing data is flagged, with a documented fallback, so a gap is not filled with an invented value.
A worked example
Twenty thousand devices, and a remote fix on the tenth ranked highest.
Round, illustrative figures for devices: they show the arithmetic, leave out the subscription, and are for you to change in the calculator.
| Input | Value | What it stands for |
|---|---|---|
| Devices in scope | 20,000 | Everything the team is responsible for |
| Value at stake per one | $150 | Service impact and support cost if a device is left to fail |
| Cost of the action per one | $8 | A remote fix or a re-provision |
| Share acted on | 10% | The devices ranked highest by value at stake: 2,000 of 20,000 |
| Modeled, USD | Pessimistic | Expected | Optimistic |
|---|---|---|---|
| Share of value at stake avoided | 10% | 25% | 40% |
| Cost to act, in total | $16,000 | $16,000 | $16,000 |
| Avoided loss | $30,000 | $75,000 | $120,000 |
| Net position | $14,000 | $59,000 | $104,000 |
Break-even is 5.3%: each action costs $8 against $150 at stake, so it has to avoid that share of the loss to pay for itself.
What exists today
Telecom and IoT exists as a catalog template and as one of the six domains compiled into the engine. It is not a live simulated fleet.
We publish a devices-supported figure only after a load test at that figure on stated hardware. If you need one, we will run it at your shape.
Bring a fleet that goes quiet
Give us a sample with gaps and irregular reporting. We will show what the engine does with them.