GERT

Tail-risk intelligence

Method / Product thesis

The average is not where grids break.

GERT is built around the operational question conventional load dashboards obscure: how close is an unlikely—but plausible—demand tail to the system boundary?

01

Observe

Official load, adequacy and weather context enter with explicit provenance.

02

Forecast

Quantile models estimate a distribution from P50 through the extreme P99 tail.

03

Stress

Scenario controls perturb physical drivers through the same inference contract.

04

Prove

Calibration gates, artifact identity and source status remain visible to the operator.

What makes GERT distinct

A single expected-load line

A visible P50–P99 uncertainty geometry

Weather conditions as the endpoint

Weather translated into capacity-tail pressure

A black-box risk badge

Risk logic, source provenance and artifact identity

A read-only monitoring dashboard

Monitor → stress → replay → validate workflow

Core decision contract

Tail demand
minus capacity
equals exposure.

Forecast

P50 / P90 / P95 / P99

Target

One hour ahead

Quality before promotion

A model is not “real” because it finished training.

Production promotion requires held-out calibration and data-quality gates. A failed candidate remains a candidate; the interface exposes fallback and stub states instead of silently upgrading their authority.

Use boundary

Decision support, not autonomous control.

GERT is a research and demonstration system. Any stub, fallback or uncalibrated output is labeled and must not be used for real grid operations. Operational deployment requires validated live integrations, governance and human authorization.