A scale that has to stay level.
The electricity reaching your socket cannot be stored in large amounts: whatever is consumed must be generated at the same instant. It is like a two-pan scale. On one side, consumption; on the other, the power plants.
The needle on that scale is the grid frequency: 50 cycles per second in Europe, 60 in Brazil and the US. If extra load comes on or a plant suddenly drops out, the needle falls. If there is surplus generation, it rises. That is why the operator chases the balance second by second.
And there is a standard saying how much the needle may wander in normal operation, something like 0.2 Hz up or down. So the tolerance the compressor needs is already written in an official document, before any of us showed up.
Quiet while the needle stays in band.
TUBE is our compressor for sensor data. Instead of storing every measurement, it draws an agreed tolerance band around the signal and stays quiet as long as the value does not leave it. Whoever opens the file later knows the frequency was there, within the guaranteed margin.
In a healthy grid, frequency drifts slowly and the compressor barely has anything to write. When a plant trips, the frequency jolts out of the band, and the compressor starts storing point after point.
It is the same reasoning as the cable that became a seismograph: how much the compressor has to speak is a measure of surprise. Balanced grid, silence. Jolt, chatter. With one detail in our favor: since the standard defines the band, the alarm is checkable against a document nobody invented for the occasion.
15.6 million measurements, 94.7% silence.
We used the frequency of the Great Britain power system, published every second by the operator (National Grid ESO), from January to June 2026: 15.6 million measurements.
Over the period, the compressor stayed quiet 94.7% of the time: the frequency wandered inside the band and there was nothing new to write. Each speaking peak in the demo chart is a real event, an imbalance between consumption and generation. The biggest took the frequency down to 49.6 Hz.
In the event window, the compressed file was 88 times smaller than the raw data, with no reconstructed measurement outside the agreed tolerance. And the compressor's speaking rate jumped 14 times above its resting level: that jump is what works as the alarm.
| What we looked at | Result | In practice |
|---|---|---|
| Data | 15.6 million | British grid frequency every second, Jan–Jun 2026 |
| Time the compressor stays quiet | 94.7% | almost the whole period fit in the band |
| Size in the event window | 88× smaller | compared with raw data, error within tolerance |
| Speaking at the event vs. at rest | 14× | the jump that triggers the alarm |
| Biggest drop of the period | 49.6 Hz | 0.4 Hz below the 50 Hz nominal |
A grid with a standard is the clean case.
Frequency is the most favorable setting for this kind of alarm: the signal is slow, the band is written in the standard and the event is sharp. Three places where the same idea applies, and one where it is not enough:
Grid operators and utilities
Keep years of frequency in little space and have every excursion flagged on its own, with a band the regulator recognizes.
Phasor measurement (PMU)
Synchronized phase meters already carry the tolerance in the international standard. It is the direct sibling of this case: synchrophasor.
60 Hz grids, such as Brazil's
The physics of the argument does not depend on the nominal value: signal in band plus an event is surprise. What is missing there is open high-resolution data.
Where it is not enough: partial discharge
In noisy signals, such as partial discharge in equipment, raw surprise alone does not isolate the event. Frequency works because it is clean.
How many times per second the current alternates: 50 in Europe, 60 in Brazil and the US. It drops when generation is short and rises when it is in surplus.
How much the frequency may wander in normal operation, written in an official document. Here it serves as the agreed tolerance.
How much the compressor has to speak. Silence when the grid is balanced; chatter when it jolts.
Where this could be wrong.
British grid, not Brazilian
We measured the Great Britain series because it is the cleanest and most open at one-second resolution. We expect the same behavior at 60 Hz, but have not measured it in Brazil.
Event is not cause
The compressor says "the frequency left the band, with proof". Saying which plant tripped, or why, is still the operator's job.
The chosen band changes the count
A tighter tolerance catches smaller jolts and also speaks more day to day. Which band to use is the operator's decision, not ours.
A clean signal helps
This is the favorable case. In noisy signals the alarm needs more than raw surprise, as we show in other cases.
← Certified telemetry · stickybit.com.br
- Data: National Grid ESO (Great Britain), system frequency every 1 s, January to June 2026, 15.6 million measurements.
- Our measurement: TUBE over the whole series (94.7% silence over the period; in the event window, 88× smaller than raw with maximum error within tolerance; surprise 14× above rest). Demo: /static/demos/gbfreq.html.
- Physical sibling: synchronized phasor measurement (IEEE C37.118, total vector error ≤ 1% as the standard's tolerance).
- Same principle: the submarine cable that became a seismograph.