We do not yet have a good way of measuring accuracy for day-to-day eye-gaze interaction. One of my collegues have summerized the current situation like this:
The current industry standard for measuring performance of eye trackers comes from research environments where eye trackers have historically been most common. Metrics are typically made in controlled enviornments where the light conditions do not change, measurements are made very soon after calibration, and the user typically does not move very much. Presenting such metrics for a dev kit intended for day to day consumer usage, gaming etc can easily be misleading, so we have chosen not to measure our eye trackers intended for consumer usage in this way.
We are actively exploring how to best measure eye trackers intended for consumer usage, and intend in the future to provide more relevant metrics on the performance of our consumer eyetrackers.