> > It's always a statistical decision. If 99.9% of such > > filtering is accurate, then the .1% that is legit, problematic > > is tough luck trade off, and must be less costly to > > handle than accepting the 99.9% that is > >crap. > > > Len, that reminds me of Fight Club and how they calculate the > cost of whether they do a 'recall' or not.
Simple actuarial analysis. It is widely used in industry today. In mechanics the old joke is you tighten the nut till it cracks, then back off a quarter turn. I have to account for the needs of my users. Because of this, some of the most heavy handed filters are not available to me without a lot of administration. Since I do not have the administration time, I have to back off, and let a larger percentage of things through. I see this in what AOL is doing. They have cranked down with a new tool. Depending on the reaction these new URL rules get, they may adjust them. But if the complaint level is low enough, they will leave them as is. If the complaint level is low, they may tighten further, say eliminate "stealth" URLs completely. And so goes the constant tweaking of anti-spam systems. --Eric
