I'm not sure why you think it's so complicated.
Blizzard has 100,00 accounts which have had player reports for botting, they know Honorbuddy has a sample of 40 specific behaviors in common, therefore if an account has performed 5+ of these Behaviors then ban. It would take very little work to do this and would probably save the investigation team considerable time. From the technical side this is decently complicated but nothing that can't be done by a single Dev.
The only reason that leads most of us to think it is not detected, is that if it were we'd be getting banned consistently and honorbuddy would be out of business. And Roboto, a sample of 1 is not overly convincing, I thought you were a statistic guy? Why would I believe the one case in thousands to be evidence of anything?
sorry but you're thinking way to small.
imagine, 5.5m online/paied accounts,
in this example, 1 account only uses one character assume 10% get reported of banning (people are also mis-using that report function, remember that!)
now carry a flag (or better a counter) across a non-centralized database (reaslms got their own database servers/blades)
taken the flag which - supposed - to be an int32 - takes 32 bytes * assume we got 550k reported players in total => 17187,5 KB alone for that flag
now add some sort of audit to that player, logging his every keystrokey, nav points and whatever not => assume it takes 250KB per 15 minutes played
assume a player bots 8 hours per day on average weather the bot's attached or not.
cary that data over 2 weels to analyze
so you got
1 player * 14 days * 8 hours per day * (250KB sample data per 0.25h *4) => we get (assumed, roughly) 110MB Data per Player
Not take those assumed 550k reported players
=> you get 60500000MB of Data to analyze => this equals 59082 GB or
~58TB data all over the infrastructure alongside with backups, regular logs and whatever.
now get that data from several datacenters together or process it on site (take 1/3 to make it easier with eu, us, asia)
so you got 20TB of data per region, now have fun alayzing that data and finding patterns within 1 week (i think one of the blueposts stated that it was a sample stopped collecting one week before banwave hit)
honestly, you cant analyze that much data reliable, nor does it make sense to even create such a pattern.
detections in other systems (punkbuster, vac2, name it) usally create honeypots, scan memory, watch for processes, hased window titles, generic thumbprints and memory pattern but they never (at least i dont know of any) ever ever analyze a players pattern) [edit: dont count that actions per minute system of SC2/HOTS/LOL - it's a completely different genere]
last but not least, taken the above example, we assumed one botter only uses one character.
now, make that 2.5 chars per bot on average => boom you got 145TB - multiply or divide that as you like
just my 2 cents