The method
Length versus angle.
Most monitoring watches how big each signal is: a voltage, a current, a frequency. That is length. Phylax-CSIM watches how the signals line up with each other. That is angle.
A power system is a machine where everything is coupled. When it starts to change, the first thing to move is often not any single reading crossing a limit. It is the relationship between readings. Two channels can each look normal while the way they move together has already shifted.
Phylax-CSIM is a chain of mathematical operators, not a trained AI model. The same inputs give the same outputs every time, every setting is frozen before a test is scored, and every flag can be traced back to the specific channels that moved.
Use what exists
Read the signals a system already records. No new sensors are needed to start.
Relate every pair
Measure how each channel lines up with every other channel, window by window.
Measure the shape
Compare that web of relationships against a baseline frozen before the test begins.
Flag and explain
Report when the shape moves, how far, and which pairs carried the move.
Focus: the electrical grid
The grid is our proving ground.
Grids are changing faster than the tools that watch them. Inverter-based solar, wind and batteries behave differently from spinning generators, and they create fast interactions that standard monitoring was never designed to see.
What standard tools average away
Most wide-area monitoring is built on phasors: a compact summary of each waveform that assumes it is a clean 60 Hz or 50 Hz sine wave and reports its size and timing a few dozen times a second. That works well for the slow behaviour of a traditional grid. It can blur or hide energy at other frequencies.
In a Southern California dataset from Caltech, Phylax-CSIM found about a quarter of the signal energy sitting in a band near 7 Hz, below the grid's normal frequency. A standard phasor estimate does not report it.
Share of measured signal energy, Southern California dataset
Case study: Eaton Canyon, 7 January 2025
Reading a fire's electrical moment from campus meters.
On the evening of 7 January 2025, an out-of-service Southern California Edison transmission tower in Eaton Canyon arced, dropped burning material into dry brush, and started the Eaton Fire. It killed 19 people and destroyed more than 9,400 buildings.
Los Angeles County Fire reached that conclusion after an eighteen-month investigation. The tower sits roughly 6 km from the Caltech campus, and Caltech publishes electrical measurements from its own distribution network. That pairing is rare: a confirmed, timed, physically understood grid event with independent instruments a few kilometres away.
The question was simple. Can the geometry of a distribution grid register a transmission fault, using only the data that actually survives after a disaster?
What the study found
The fast data did not exist.
Phasor and point-on-wave records, the fast waveform measurements that would show a fault forming, are not kept. Utilities retain them for seconds around a trip, not for the hours before. This shaped the whole study, and it is the most transferable lesson.
What a university keeps was used instead.
Caltech's distribution meters report roughly one sample every ten seconds. From 22 meters the study built 60 power-factor angle channels and asked how every pair of channels moved relative to each other: 1,770 pairs in all.
The method saw the fault.
At 18:11 PST, Southern California Edison recorded a fault on a line more than five miles away that put a brief surge through the four live lines over Eaton Canyon. The geometry statistic hit its largest value of the entire study inside exactly that ten-minute window, larger than anything across ten ordinary baseline evenings. That is the strongest result this baseline can resolve (rank p = 0.0058).
It did not show a precursor, and we say so.
The test for an earlier warning before the fault came back null. Our analysis shows why: a slow, spread-out change is diluted across many pairs, where a sharp fault concentrates in a few. Seeing a precursor would need faster data, not a looser test.
Everything was fixed in advance.
Channels, windows and baseline were set before the event evening was scored. The result can be rerun by anyone with the same public meter data.
Case study: Iberian Peninsula, 28 April 2025
Half an hour of warning signs, then seconds of collapse.
At 12:33 CEST on 28 April 2025, mainland Spain and Portugal lost power. It was Europe's most severe blackout in more than two decades.
The ENTSO-E Expert Panel, 49 members drawn from grid operators and regulators across Europe, published its final report on 20 March 2026. It found no single point of failure. The blackout came from many interacting factors: oscillations, gaps in voltage and reactive-power control, differences in how voltage was regulated, rapid output reductions and generator disconnections in Spain. Together they drove fast voltage rises and a cascade of generators tripping off to protect themselves.
Two episodes of oscillation came in the half hour before the collapse. The Panel found they did not cause the blackout directly, but they shaped the state the system was in when it failed. The measures used to damp them also pushed voltages higher.
Why this event matters to us
A change in how the network's parts move together, unfolding over roughly 30 minutes before a collapse, is exactly the kind of signal cross-channel geometry is built to watch. The day was also geomagnetically quiet, which points the search inward, at the grid itself.
Where we stand, plainly
Our current Iberia analysis rests on a single waveform node, at Málaga. One node is not enough to claim a precursor, and we do not claim one. We have identified the Málaga and Riga sites as an electrically meaningful pair, since Riga joined the Continental European synchronous area in February 2025.
What would settle it
Continuous waveform records from measuring points across Spain and Portugal, covering 28 April 2025 and the weeks before it. With enough sites, a frozen baseline built from ordinary days can be scored against the event day, the same way the Eaton study was run. We are actively seeking those records.
For utilities and measurement networks
The bottleneck is data, not math.
Both grid case studies hit the same wall. The measurements that would show a change forming are rarely kept long enough to study. We are looking for partners who hold them.
We work under non-disclosure, scope agreements to your data, and publish only what you clear.
- Three-phase voltage and current at the same pointThe cheapest step with the largest gain. It lets each site be read against itself, as at Eaton.
- Density near the eventSeveral measuring points close together, so a change can be seen spreading from pair to pair.
- Point-on-waveFull waveform records, kept continuously, for the hours and weeks before an event rather than seconds after it.
Talk to us.
Briefings for grid operators, insurers and research partners. We will walk through the Eaton and Iberia work line by line.
slayne@syncreticai.com