The validation uses FCR-proxy events, not operator-declared instabilities. How do I know this generalises?
FCR-proxy events (|Δf| ≥ 0.10 Hz) are a conservative, objective threshold derived directly from grid frequency data. They are not a substitute for operator-declared events, and cross-validation against operator records is in progress. What the validation establishes is that CANAREON detects the precursor dynamics preceding these threshold crossings — consistently, across two independent grids, with meaningful lead time. Generalisation to other grid configurations and event definitions requires further independent validation, which is the purpose of pilot engagements.
CANAREON is a single-researcher organisation with an unpublished preprint. What is the dependency risk?
The BCI engine is a locked, documented mathematical system — not a proprietary black box. The canonical equations, validation methodology, and implementation are fully documented in the preprint. Any competent research team could reproduce the results independently. The intellectual property is the framework; the organisation is the delivery vehicle. The preprint is submitted to SSRN and will be available for independent review upon DOI assignment.
What does a pilot engagement actually involve?
See the How a Pilot Works page for a full breakdown. In summary: a briefing call to assess fit, a data assessment on a sample signal, a structured pilot run (30–90 days), and a formal review. Data is handled under NDA. No integration with live systems is required for the initial pilot.
Is CANAREON safe to use in operational settings?
No. CANAREON is experimental research software. It is not safety-certified and does not replace operator judgment. It is designed as an additional layer of visibility — an early warning signal that operators can choose to act on, not an automated control system. Deployment in safety-critical operational settings requires independent validation and appropriate regulatory clearance for the relevant jurisdiction.
Why doesn't CANAREON use machine learning?
See the Why Not AI page for the full argument. In short: ML models require training data, are subject to distributional shift, produce non-auditable outputs, and face significant regulatory barriers in safety-critical infrastructure. BCI produces closed-form, auditable, deterministic outputs that are reproducible across domains without retraining.
What is the difference between BCI and BCI-M?
BCI (Burden-Coupled Instability) is the canonical two-variable model validated on power grid data. BCI-M is a three-variable extension that adds a Vulnerability Index — a memory of accumulated stress history. BCI-M is used by the DECIDE module for human decision environments, where prior stress history shapes current decision capacity. The BCI canonical engine is unchanged by BCI-M; it is implemented as an external wrapper only.