Since January 1st, 2015 ChessTempo Mixed-Mode has been my primary tactics trainer. There are two reasons. The first reason is simply the high quality of the problems--the problems are curated very well by somebody on the ChessTempo team. The second is that Mixed-Mode includes both offensive (mate or move and win) problems and defensive problems. You never know what you're going to get and this uncertainty allows me to treat my daily Mixed-Mode training sessions as game simulation training, if you will. IE, it is quite probable one would spend 3-6 minutes or more on a critical position in an OTB game, checking every nuance of a move before committing to it.
I strongly believe my consistent Mixed-Mode training has significantly improved my OTB chess strength over the past 18 months, taking my OTB rating from 1530-1751. I'm confident this is beyond where it would have been if I were only using the more traditional (offensive-only) tactics training tools (which I also use in a "woodpecker" or "saltmines" training style, where I quickly train the same problems over and over, going for memorization of patterns).
ChessTempo's problem selection algorithm serves up a wide range of problem ratings centered around one's own current rating. I'm currently 1710 in Mixed-Mode and have recently been served problems rated from 1500 to over 1900. This wide rating range also obfuscates the solution because one cannot assume the solution that instantly pops out is the correct one--it could be rated several hundred points above one's current level, meaning there are likely to be some tactical nuances in the solution that need to be calculated. Or, it is one rated two hundred points below one's current rating, meaning that, yes, that first intuitive solution that popped into one's head is likely the correct one.
When I use Mixed-Mode, I solve for accuracy. I want to get 100% of the problems correct and I usually take 3-6 minutes to solve each problem, though rarely over 10 minutes. My actual percentage correct is 68.94%, owing to the auto-adjusting algorithm that the site uses to serve problems.
Here is a problem that is typical of Mixed-Mode that I failed to solve this morning.