Silktide records all visitor data, but when showing aggregate values we sometimes use samples.
For example, say a website had received 10 million views during a period, and you were applying complex filters for location, browser, and so on. Checking all 10 million views could result in unacceptable slow performance, so in such cases, Silktide might use a smaller sample and extrapolate from that. These samples are large enough that any error is almost guaranteed to be <0.5%.