Reading about testing round trip latency to determine how accurate the audio driver is reporting.  Seems as simple as recording or loading a simple audio clip, then routing the output of the interface to an input of the interface and record to a new track and observe the delay in samples.  
 
Ok, I can do that...but doesn't the latency change as a project evolves, adding and removing tracks, muting and playing back multiple tracks, recording 10 or more tracks at once..and etc.?  I don't see how I can enter a static number for latency offset while latency remains dynamic.
 
I like the idea of testing and making latency calculations more precise, but I thought the audio driver was continuously handling the dynamics of latency as the project evolves.  Am I way off on this?  

I use the manual offset to adjust the ASIO reported latency. Every interface I have used needed an offset. Centrance provides a utility to perform test. The latency reported by the driver and the offset are static values. The only latency that changes on a per-project basis is the Plug-in Delay Compensation. PDC has nothing to do with the driver and offset values.

The audio driver attempts to operate with the latency you set. If the project demands more from the computer than the computer can provide as regards preventing the buffer from emptying you get crackles, drop-outs etc. Until you are asking more of the computer than it can handle adding audio tracks won’t make much if any difference. Sonar, like other DAWs, does not handle the buffer setting “dynamically”, it simply attempts to pass data through the driver.

Plugins can add latency if they need more time to do their thing than the buffer and computer can handle. Some plugins have a fixed amount of latency which sometimes can be quite high (in a few cases as high as a second or more). Sonar, like pretty much every DAW, compensates for the time required by extending the latency time to take account of them.

The manual latency offset is there to deal with the situation that the ASIO driver doesn’t report the latency correctly to Sonar or the interface/driver places audio in not quite the right time-slot for some reason. The latency isn’t reported correctly by some interfaces because they contain a “hidden” safety buffer with a fixed value in time/samples, which can sometimes be quite significant. If that field needs an entry in it a single fixed amount will usually work for all buffer settings.

A good ASIO driver will report to Sonar the needed offset and most of the times this will be within a few samples. It's so close as to not worry about it.
Being 10" from your mike will do the same thing. 
You certainly wold not want to use anything but ASIO mode. 
Here is one of my loopback tests with my Tascam and Scarlett.  

Given how easy the the fix is, I don't see why one would be satisfied with an incorrect latency value. This is even more important when using multiple interfaces as some under-report and some over-report latency. I would imagine over time, all of us use more than one interface and having the correct latency setting insures the clips are lined up correctly regardless of the interface used.

Thanks guys.  I feel a lot better about this now.  It sounds like I don't have much to lose by testing the latency and at least compare it to what my driver is reporting.  If it's off, I might as well fix it. 
 
And it doesn't sound like interface latency settings are a dynamic evolving parameter like I thought - thanks for clearing that up.  Any dynamics in latency associated with other things appear to be dealt with by the DAW, so that's comforting.
 
Latency really becomes an issue for fast percussive instrumentation and I want to aim for the best precision that is reasonably possible.  Was reading a helpful thread in another forum, that I am not a member of, and the guy was setting his offset based on a microphone sitting next to his ear, aimed at the monitors he would be listening to when tracking guitar.  The difference between the speakers and his head adding 1.8 ms - and he figured that in. 
 
That might be a little much for me, but I admire the commitment, ha ha.   
 
 
 
 

I would think 1.8 ms is well below the threshold of what we should be worrying about. I start to hear latency at around 7ms. I guess it comes down to wanting music that is perfect, I would think perfect music would belong in an elevator :) 

I encountered weird latency on Sonar. I thought it's some temporary system issue, but after a few days I checked my 6i6 in other DAWs and standalone programs - all perform as usual, that is virtually unnoticeable latency with low buffer settings (32 to 128 on my 6i6). Sonar not only monitors sounds with a huge latency, but also records both audio and MIDI with a huge offset. Any clue on how to fix this?

Cactus Music
I would think 1.8 ms is well below the threshold of what we should be worrying about. I start to hear latency at around 7ms. I guess it comes down to wanting music that is perfect, I would think perfect music would belong in an elevator :) 

Well, I can't speak for others, but in my case it's not about music being perfect or I would just record MIDI notes from an electronic kit and quantize.  Instead, it's this weird sensation of the sound of my drums slightly behind the feel of them.  I worked through several songs like this and I could never quite accept the delay between hitting and hearing.  It's not that noticeable, but it's just enough to never feel "on" and in the pocket when listening through the isolation cans.  Electronic drums felt the same way.   

You have changed the subject from recording audio to recording MIDI. The manual offset affects the location of recorded audio on the timeline.
 
When using an ASIO driver the offset is intended to correct the reported driver latency. As I mentioned above some drivers over-report and some under-report latency. My current interface over-reports latency by 21 samples. My last interface under-reported by 76 samples.
 
I suppose the setting could be used any way you wish but I have never heard of anyone using it to address the distance between a listener and the monitors. Not sure that is a good idea.
 
FWIW, here are a couple of threads about the setting
http://forum.cakewalk.com/FindPost/1301345
http://forum.cakewalk.com/FindPost/3178396