Product · Methodology

How Flexgy decides, and when it does not

Four steps in order, each with an explicit threshold. If any of them fails, the decision falls back to the DSO’s conservative physical rule.

Calibrated margin
Flexgy measures how wrong your engine has been in the past at that node and under which conditions. From those errors it computes a margin for the risk level you choose (P50 to P99), instead of a fixed figure for every case.
Technique: conformal prediction on the engine’s errors.
Never-seen state
The margin only holds if the grid resembles the calibration period. Flexgy measures how far the current state (temperature, demand, generation, topology) is from anything it has seen. If it is too far, it does not trust the margin.
Technique: out-of-distribution detection with a threshold set in advance.
Physics check
Before recommending anything, the limit is checked against the security criteria: thermal loading of the elements, voltages within limits and the N-1 criterion.
The same criteria set by the POD1 operating procedure.
Decision and record
If everything passes, it acts. If confidence falls short, it limits just enough. If the state is new or the physics does not close, it abstains and applies the conservative rule. In all three cases an Evidence Passport is produced.
Fixed, versioned rules, not judgement of the moment.

Node N-12 · 45 kV

Example substation · 3 Type 2 flexible permits

Type 214 Jul 2027 · 16:40

Candidate limit

14.2MW

Grid engine

Calibrated margin

—

No valid calibration

Grid state

41.2 °C

Out of domain

Decision

11.9MW

ABSTAIN · physics rule

Last 24 hMW

Engine candidateCalibrated P95 bandAuthorisedGuaranteed firm

Decision log

16:40	ABSTAIN	11.9 MW	000419
16:25	ABSTAIN	11.9 MW	000418
14:05	ABSTAIN	11.9 MW	000391
13:50	ACT	13.4 MW	000390
11:15	ACT	13.1 MW	000377
21:30	LIMIT	10.8 MW	000342

Temperature 41.2 °C: outside the calibration range (max 38.6 °C). The DSO’s conservative physical rule applies.

Open passport FX-2026-000419

The four steps on one screen: the engine candidate, the calibrated band, the never-seen state and the recorded decision. Illustrative screen with synthetic data.

Assumptions and limits, in writing

Calibration assumes the future resembles the recent past. That is why step 02 exists: when it does not, Flexgy says so and steps aside.
Flexgy does not fix a poorly modelled engine: it detects it, because its past errors widen the margin.
Abstaining is not a failure: it is the right answer when there is not enough evidence. It is measured and reported.
The risk level is chosen by the DSO, not by Flexgy.

How it is measured

The metrics we agree before every pilot, with success and stop criteria.

False-safe decisions: Times a limit was authorised that reality would not have withstood. This is the metric that rules.
Capacity utilisation: How much flexible capacity is used compared with a fixed margin.
Abstention rate: How often Flexgy falls back to the conservative rule, and why.
Calibration: Whether the chosen risk level holds in practice (e.g. P95 ≈ 95 %).
Latency: Decision time, compatible with real-time instructions.

Status

The methodology is implemented and validated on synthetic data. Figures on a real grid will only exist after a pilot, and we will publish them with their context.

Want to review the methodology with your technical team?

Request a technical session

Calibrated margin

Never-seen state

Physics check

Decision and record

Assumptions and limits, in writing

How it is measured

Status

Keep reading

Platform

Evidence Passport

Pilot programme

Want to review the methodology with your technical team?