Some observations from that report: When they tested the filters they reported "less than 3.4%" of innocuous URLs were mistakenly blocked. They say this is too

Some observations from that report: When they tested the filters they reported "less than 3.4%" of innocuous URLs were mistakenly blocked. They say this is too high (no kidding!), but that "blocking rates of below 2 percent would be considered low" and thus acceptable. They also considered performance degradation of 0-10% to be "negligible", and 10-20% to be "minimal". The performance results looked incredibly noisy to me, but they don't report variance or any kind of statistical significance, so I don't think their results mean much.