SONAR doesn't really have a buffer of its own. The ASIO buffer is set in the driver/firmware of the interface via the ASIO control panel, and SONAR uses that value to read/write from/to the driver.
 
The total roundtrip shown in SONAR is whatever the interface driver reports. It's not uncommon for the driver to omit some or all of the hardware latency (i.e. D/A/D conversion time and USB/FW/PCIe bus latency). CEntrance and External Insert are actually 'pinging' the loop and measuring the elapsed time so they pick up all the HW latencies.
 
I don't know why the EI reports a value that's one buffer short, but it does work correctly in real time, and when bouncing/freezing.

I've just left a link to the online documentation regarding External Inserts. The manual explains it much better than I can.

http://www.cakewalk.com/D...mp;help=Mixing.33.html

Hope it helps.

Last night I used CEntrance to measure all buffer settings that I normally use for tracking.  The 'other' buffer setting in my Hilo is 'USB Steaming Buffer". 
 
For my setup, taking CEntrance number - Sonar asio = 154 (+/- 1 sample) consistently.
 
The -102 offset condition I mentioned earlier was at a setting (asio and usb combination) that the HiLo doesn't recommend, although it's an accurate offset, so for practical purpose I'll not use this setting.
 
In summary I've confirmed that:
- The un-reported HW latency is a constant value at any asio/usb buffer.
- Sonar asio reports correctly the latency it received from the HW.
 
I've never thought that I need to 'calibrate' the HW as per the documentation said it should be very small or perfect. So all this times I have tracked with this latency.  This exercise leads me to think otherwise.  I think that measuring your HW latency is a must regardless.  CEntrance really makes this measurement easy and accurate.
 
@ mudgel - thx, I've read that page many times.
 

My recent experience is mostly with PCI(e) interfaces which I like for the very low bus latency. My old E-Mu 1820m required only a 22-sample offset. The MOTU I run now requires 44 because the driver under-reports more, but it has a lower net hardware latency than the 1820m. It adds only 89 samples total to the I/O buffer latency, and that 44-sample offset is all converter latency. It goes away if I use digital I/O.
 
USB interfaces are all over the map in this regard, but I think Lynx are pretty good.
 
Glad to hear you got it all sorted in any case.

One more thing.  Once you have set the asio manual offset, in addition to bounce and freeze, you can record a track with EI by set the In of the new track to the same port as the EI' return (if your HW has mixer that allows the route).  The recorded clip would just lined up perfectly.