CANAREON

Detect instability before failure.

Minutes to hours of advance warning across real systems. Validated on national power grids. Built on deterministic mathematical models, not AI.

CANAREON identifies instability as it forms, enabling earlier intervention and reducing the risk of system failure.

Mechanism

How CANAREON Works

CANAREON measures how systems respond under stress — not just the conditions they operate in.

Capacity Reserve
Capacity
Tracks the system's remaining capacity to absorb disruption without destabilising.
Instability Signal
Instability Signal
Identifies instability as it forms, before it becomes visible in standard monitoring.
Stress Accumulation
Vulnerability Index
Captures how prior stress history shapes current system resilience.

These signals combine to identify instability as it forms, before failure becomes visible.

From Signal to Decision

CANAREON translates system behaviour into decision-ready outputs, including system state, time to impact, and recommended action.

State
Stable to Critical
Time
Minutes to hours
Action
Monitor, intervene, respond
Validation

Validated on Real Systems

CANAREON has demonstrated early warning capability across real-world systems, identifying instability before failure becomes visible.

Power Grid Validation

Tested on national power grid data, CANAREON identified instability signals in advance of observed system events.

  • Up to 165 minutes of advance warning
  • Validated on Finnish national grid data
  • Consistent signal detection prior to stabilisation events

Validation is retrospective and based on proxy ground truth (frequency containment reserve events).

Cross-Domain Consistency

Similar instability patterns have been observed across multiple systems, supporting the generality of the underlying model.

  • Traffic systems
  • Energy systems
  • Complex operational environments

Instability emerges before failure. CANAREON makes that signal measurable.

Full methodology and validation results →

Early Warning

Minutes to hours before failure is visible through conventional monitoring. Fold Proximity and Lead Time quantify how much intervention time remains.

Multi-Domain

One engine across validated domains — power grids, ecological systems, AI training instability. The same mathematical kernel. No retraining.

Deterministic

No black box. Five interpretable states from S0 STABLE to S4 COLLAPSE. Every output is traceable to a closed-form equation.

Decision Layer v1

Five operational states. Deterministic.

Continuous instability signals translated into actionable classifications — in real time.

🟢
S0
STABLE
Normal operations
🟡
S1
EMERGING
Prepare: monitor and ready
🟠
S2
ESCALATING
Intervene: reduce load
🔴
S3
CRITICAL
Respond: immediate action
S4
COLLAPSE
Recovery: stabilise
Origin

The Instrument That Preceded the Instrument

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 →
Validation

Validated on two national power grids.

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 Briefing

Ground 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.

Who It's For

Three domains. One engine.

GRIDValidated
Power Grid Operators
Advance warning before frequency instability. Validated on two independent national grids.
29 min median lead time (NESO GB)
TRAFFICAdvancing
Transport Managers
Advance warning before congestion cascades in road networks.
98.5 min median lead time (METR-LA)
DECIDEPreview
Leadership & HR Teams
Advance warning before decision capacity failure in high-pressure environments.
BCI-M framework locked
How It Works

Signal → State → Action → Outcome

Signal
Sensor time-series data fed to the BCI engine.
State
Continuous outputs mapped to S0 – S4 operational states.
Action
State-specific recommended action surfaces to operator.
Outcome
Minutes to hours of lead time before visible failure.
Context

Why Now

Grid complexity is rising
The energy transition is adding renewables variability, interconnection stress, and rapid inertia changes to grids designed for baseload stability. Legacy threshold monitoring was not built for this operating environment.
Oversight requirements are intensifying
Regulators and system operators face growing pressure to demonstrate proactive risk management — not just reactive incident response. Early warning is becoming a governance expectation, not just an operational advantage.
The mathematical framework is ready
BCI translates complexity science directly into operational early-warning outputs. No AI training, no black box, no distributional shift. The same kernel validated on one grid runs on another — unchanged.
Why mathematical instability theory, not AI? →
Live
Live — updated continuously

Real-Time Grid Instability — Now

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 →