Open validation protocolStatus: data collection not yet published
Predicted versus measured

How the ArcticSoak model will be tested.

The model should earn confidence by publishing residuals against real installations—not by repeating its own assumptions.

No validation result is claimed yet. This page publishes the protocol and minimum evidence standard before the dataset exists.

Required installation fields

Minimum observation window

Seven complete days is the minimum for a provisional observation; 30 days spanning normal use is preferred. Startup cooldown must be separated from hold-temperature consumption.

Published outputs

For each accepted record, ArcticSoak intends to publish predicted kWh, measured kWh, absolute error, percentage error and enough configuration detail to explain the comparison without identifying a private address.

Calibration rules

Model constants will not be changed to fit a single installation. Recalibration requires a diverse sample, out-of-sample evaluation and a model-version increment. Removed observations and exclusion reasons will be documented.

Planned sample

The initial target is 40 installations across multiple climate zones and setup types. Until that threshold is reached and results are published, ArcticSoak labels model outputs as unvalidated planning estimates.