Data source. The address is geocoded, then the tool reads an aerial elevation model (a digital surface model derived from aerial lidar survey data, requested at 0.1-meter pixel size) together with a building mask and aerial imagery for that location. The measured boundary is the one the visitor confirms on the imagery before measurement runs.

Method. The engine fits planes to the elevation data inside the confirmed boundary, classifies the resulting facets by pitch, and derives ridge, hip, valley, and eave lengths from the fitted geometry. Area splits at 3/12: steeper facets report as sloped roof, 2/12 and under as low-slope. The printed report states the survey flight year and lidar point density for the property, because the answer is only as current as the flight.

Confidence tiers. Tier A and B results include the supported edge network; when the underlying data cannot support edges honestly, the tool falls back to tier C, which reports area and squares only rather than showing low-confidence linework.

What it is not. An aerial estimate, never a surveyed takeoff, a price, or a contract quantity. Tree cover, construction newer than the survey flight, and gaps in lidar coverage all reduce what can be measured, and the report says so. A documented on-site measurement is the final word for scope and material orders.

Method. The photo is resized on the visitor's own device, which also strips camera metadata such as GPS location, then sent once to an outside image service (OpenAI) to render the selected material onto the roof in the photo. The color choices are the published palettes of named manufacturers, entered from their printed color sheets; the result returns to the browser and is not stored on our servers.

What it is not. A design concept, not the finished roof. Lighting, photography, and screens all shift color, so the concept narrows choices and a physical sample in daylight confirms them.