daroz - 3:35 pm on Apr 18, 2003 (gmt 0)

Assuming they are using a fraud prevention algorithm, my philosophy with verifying the click throughs is that close is good enough -- what do you think?

Basically, yes. There's one more thing I do with my referral tracking. First, it's all done in real time on-line via a backend database. (No logfile parsing). I have a cronjob that queries the DB for the last 8 15 minute period referral stats by source (adwords seperate from banner ads etc) and compares it against the same periods 1 day and 1 week ago.

I have some threshold values setup so that if there is a sudden jump in referrals either an e-mail is sent or I get an IM notification with the stats. (I actually use Jabberd to provide web-based chat on this site, so it was an easy extension).

In other words, if someone goes on a clicking bender I'll know within 15-30 mins.

Do you guys actually match them click for click with your logs?

I tried, and found I'll usually get pretty close on slow days. The thing that kept thowing my numbers off was the fact that Google measures a day based on Pacific Time (now GMT-7 IIRC) while I measure on Eastern time. Once that was fixed I found the margin of error to be less then 1% when there were no multiple-clicks in the sample period. It's an important distinction: the 'error' rate will change when the fraud detection kicks in.

After screwing around with the fraud detection on a $.05 keyword (I paid for) I'm confortable enough that I know what to look for on my side to catch it so that I can let Google be Google and not agonize over differences between their numbers and mine.