CANAREON measures how systems respond under stress — not just the conditions they operate in.
These signals combine to identify instability as it forms, before failure becomes visible.
CANAREON translates system behaviour into decision-ready outputs, including system state, time to impact, and recommended action.
CANAREON has demonstrated early warning capability across real-world systems, identifying instability before failure becomes visible.
Tested on national power grid data, CANAREON identified instability signals in advance of observed system events.
Validation is retrospective and based on proxy ground truth (frequency containment reserve events).
Similar instability patterns have been observed across multiple systems, supporting the generality of the underlying model.
Instability emerges before failure. CANAREON makes that signal measurable.
Minutes to hours before failure is visible through conventional monitoring. Fold Proximity and Lead Time quantify how much intervention time remains.
One engine across validated domains — power grids, ecological systems, AI training instability. The same mathematical kernel. No retraining.
No black box. Five interpretable states from S0 STABLE to S4 COLLAPSE. Every output is traceable to a closed-form equation.
Continuous instability signals translated into actionable classifications — in real time.
Before electronic gas detectors existed, miners carried canaries into tunnels. The canary's sensitivity to toxic gases created a window of detection — time to act before the invisible became fatal. CANAREON applies this principle mathematically: turning system response into a leading indicator of risk.
Read the full story →Validated on two independent national power grids. Median 29-minute advance warning on NESO GB (334 days, 8,245 events). Median 165-minute advance warning on Finnish grid (365 days, 490 events). 82.9% of NESO GB events detected ≥10 minutes before threshold crossing. AUC-PR: Finnish grid 0.483, NESO GB 0.4443 — difference 0.039.
Retrospective · proxy ground truth (FCR events) · not operator-declared instabilities
Request BriefingGround truth: FCR-proxy events (|Δf| ≥ 0.10 Hz); not operator-declared instabilities. Results are not safety-certified and do not replace operator judgment. Research status: experimental. See science page for full methodology and limitations.
The CANAREON pipeline is running live on FinGrid national frequency data. This is not a simulation. Instability signals are computed and updated in real time.
View Live Demo →