Skip to content
grid-diamondcutter

How the methodology works

This page is a short orientation. The full methodology lives in the repository.

The shape of the project

The repository contains:

  1. A toy power-grid simulator (power_grid_sim.py). Small enough to read in one sitting. Runs in one second. Produces a complete demonstration of the method.
  2. A substrate-plugin contract. If you replace the toy with your own grid model behind that contract, every measurement in the project re-runs against your model.
  3. A collection of measurements, each a separate Python file. Each measurement is a four-part unit: prediction, condition that would prove the prediction wrong, code, result.
  4. A set of self-checks. Each thing the project commits not to claim has a small checker that scans the repository for that forbidden claim. The checkers run on every change.

What “cycle-walk” means

The cycle-walk pattern asks: under realistic operating cycles (winter peak, summer cooling, fault-then-restoration, hurricane recovery), what does this grid’s behavior look like, and how much of the cost is recoverable inefficiency that better routing could avoid?

The full theory document is at docs/methodology.md. A translation layer for engineers coming from PYPOWER / pandapower / OpenDSS / MATPOWER is at docs/for_grid_engineers.md. The polyphonic meta-sim layer is described at docs/meta_sim.md.

What “pre-registered” means

Before any measurement runs, the prediction it is going to make and the condition that would prove the prediction wrong are written down and committed to the repository. This is the same discipline clinical trials use: a researcher cannot quietly change their prediction after seeing the answer, because the original is on the record.

The full pre-registration document is at PREREGISTRATION.md. The phase-2 extension is at phase_2/PREREGISTRATION_PHASE_2.md.

Why both pass and fail are kept

Hiding failed predictions makes a project look like it always works. Most real projects do not always work. Keeping the failures on the record is how a reader knows the passes are honest.

In this project, about six out of ten predictions failed. That is within the healthy range. A project where every prediction passes is flagged for review automatically.