My experience with mirror compasses is that 2 degrees are typical for repeatability.
Keep in mind that the 8099 and 15TDCL are indirect reading, so not only do you have optical issues (boxing the needle), but also
the manipulation of the circle as well.
Toss in some vertical misalignment too.
True prismatic's (Sirs G150,PG M-73 or Suunto KB14, KB-360') can be read repeatability to about 1/2 degree.
Never mind any claims to "1/4 degree" or "1/6 of a degree".
You would need a surveyors compass and tripod and then still use repetition and averaging to get 10 min of arc.
That assumes that any local attraction can be compensated by shooting both sides of the line.
Anything hand held is really difficult to read to 1/2 degree, nearest degree is realistic.
A number of caving organizations use true prismatic compasses for underground surveys and have published
a lot of experimental data on accuracy and repeatability.
It surprised me to learn that most of these still had routine errors of several degrees (offset).
http://www.chaos.org.uk/ has a number of articles on the topic.
The old lensatic compass is actually specified as 2 degrees of accuracy and 1 degree of repeatability.
Most people are fooling themselves if they are getting better results with either a lensatic or a mirror compass.
Don't get me wrong land navigation can be done with 2 to 3 degrees of accuracy, it's surveying that requires
better results, that's why the military mortar and artillery units need the M2 (military version of pocket transits)
or the G-150,M-73, even then these are backups only.
Try going to a field and mark an irregular shape with cane poles or other markers, make sure the distances are at least 100 FT.
Now walk from point to point and shoot both forward (forward sight the next pole) and reverse (back sight to the last pole).
Then reverse the process, record all the results and then plot them out.
The results will speak for themselves.