The interpreter's 0-1 confidence that this localization is correct.
Short human-readable label for the element (e.g. 'Sign In button').
Pixel X of the element's centroid in the screenshot's coordinate space.
Pixel Y of the element's centroid in the screenshot's coordinate space.
One UI element the visual interpreter localized in the screenshot — a label plus the pixel centroid the voice agent can feed straight into
browser_Click(x, y). Coordinates are in the SCREENSHOT's own pixel space (top-left origin), which equals the live browser viewport.