Withstand

The building is finished. The model is what's holding up energization.

Withstand turns facility settings and one disturbance record into a PERC1 package your interconnection reviewer can accept — with the evidence behind every number.

Built for the engineers who submit large-load packages to Dominion and PJM, and the reviewers who have to sign them off.

Facility envelope / systems cutaway
Measured versus modelled active power through a voltage disturbanceActive power holds at 1.0 per unit, drops to roughly 0.1 per unit when the load ceases at 0.5 seconds, holds through the cease interval, then ramps back to 1.0 per unit after reconnect at 1.6 seconds. The modelled trace follows the measured trace closely.1.00.5001234P (pu)t (s)CEASERECONNECTmeasuredmodel · calibrated, holdout-scored
Fig. 1 — Active power at the point of interconnection through a voltage sag. The cease and reconnect transitions are the behaviour a reviewer asks you to substantiate, and the interval between them is where most submitted models diverge from the real facility.
withstand profile .

Start with the free screen. It runs locally, lists what the governing profile asks for against what you already have, and sends nothing off your disk.

Of 33,282 MW of operating data-center load, 25,504 was submitted with no explicit dynamic model at all. That held right up until reviewers started asking.

NERC Level 2 aggregated report, March 2026

Every deficiency cycle costs weeks. The building is already standing, and it earns nothing until the model clears review.

Settings sheets and one disturbance record become a PERC1 package in PSS®E format — with the record behind every number a reviewer can check.

And when your records cannot support a number, the package says so instead of inventing one. That refusal is why the rest holds up.

We measured which PERC1 parameters a disturbance record can actually support, on synthetic records with known ground truth — including the case where a fit reports full rank and is still wrong. Read the technical note.

Run the screen yourself.

Two commands, both local. The client never uploads your model, your evidence, or your records — the paid operations are separate and explicit.

  1. withstand init . --name my_facility --profile dominion-2026-04-01

    Writes three plain files: the manifest, the model source, and an identity lock. Nothing is uploaded.

  2. withstand profile .

    Scores your project against the thirteen rows of the governing Dominion profile — satisfied, needs attention, or manual review — each with its authority, effective date, and source.

Builds go out to design partners first. Ask for one and say which platform you need. The compute engine stays server-side and is never part of the download.

The screen is free. The engagement is fixed-fee, scoped to one facility.

Quoted against one deficiency cycle, not against software — and not billed hourly.

Readiness screen — what the governing profile asks for, and what you have
free
Identifiability screen — which claims your records can actually support
free
Settings-locked calibration with held-out validation
included
PSS®E and PowerWorld export with provenance sidecars
included
Signed evidence packet, verifiable offline without us
included
Gap analysis and commissioning test plan, when the records refuse
included

The readiness gate comes first. If your records cannot identify the behaviour the reviewer is asking about, you get the gap analysis and the commissioning test plan that would fix it — not a fitted number nobody can defend.

Tell us about the facility.

One reply from a person who has read it, usually within a day. If the project isn't a fit we'll say so rather than book a call.

Useful to include

  • Facility, delivery point, and the target submission or energization date
  • The simulator and exact version your reviewer names
  • Whether you have a disturbance or commissioning record, even one
  • Any deficiency comments you have already received