Aero testing protocol tuning

I've recently been playing around with some aero testing protocols.
In particular, I've been trying to use the "Chung method", which relies
on computing Virtual Elevation (VE) based on the inputs from a power
meter and your speed.  Ideally, if you have a good testing protocol,
the can give you results that are similar to having a wind tunnel.  That
means you can get feedback on position and equipment changes without
visiting a wind tunnel (and paying the big bucks).  This sort of field
testing does seem to require a bit of care.  I've been working on getting
the hang of what are the important bits so far.  There are some folks who
use a spreadsheet, but for now I'm using Aerolab in GoldenCheetah.

My first try was back in Feb, where I had all sorts of problems:

1. It was really windy.  Bad bad bad.
2. There were passing cars. Double bad.
3. I was varying equipment before verifying the protocol.
4. I was using GPS for speed.  Yuk.
5. ...

What was cool was that it still seemed to show some coarse
differences in CdA between the large scale equipment changes.
At the time, I was using my powertap, so I couldn't really test
my full race setup since I race with H3Cs.  I got some feedback
on slowtwitch which suggested I needed to refine my field testing
before going further. http://forum.slowtwitch.com/gforum.cgi?post=3508712#3508712

Recently, I got myself a quarq (very nice) and decided to give
it another shot now that I can test my actual race setup. 

Here's what that looks like by the way:

For the new testing, I changed a bunch of things:

1. Picked a new venue, a half-pipe close to home where I can turn around slowly on the uphill ends.
2. Added a speed sensor and wheel magnet.
3. Upgraded my Garmin 705 to new firmware (3.3)
4. Varied the power quite a bit from lap to lap.

My first decent test looked like this:

Here the elevation actually matches reasonably from the Garmin for the last 5 laps or so.
It isn't entirely clear what was going on the first few laps, some combination of not hitting
my marks, cars passing, not keeping an identical position, etc.  The raw data can be
found at:  http://connect.garmin.com/activity/110917690

I tried again a couple of times today.  The equipment was identical to the day before
(different clothes, helmet), but the idea at this point was to see if I could generate
repeatable runs.  The first time I tried, the wind was perfect, but I was getting a
little bit of car traffic (messing up the "wind") and some foot traffic that was
messing up my lines.  I eventually gave up and rode around for a while and then
came back later.  Here are those two attempts:

 

 

Clearly the second try was much better (although the first was still better than my previous attempts).  The second to last lap has a pretty clear glitch.  I think that was a truck going by.  The route was pretty isolated otherwise (there may have been a car passing the other way on one or two of the other laps).  Unfortunately, the altitude was glitchy, so I ended up zeroing it out.

Raw data at: http://connect.garmin.com/activity/110919793 and http://connect.garmin.com/activity/110920056

So what's next.  The suggestions I've gotten so far:

1. Mark the turn around carefully with chalk (point of entry and apex)
2. Calibrate the power meter.
3. Use more than one wheel magnet.
4. Weigh things carefully with a good scale.
5. Get a Kestrel 4250 (portable weather tracker).

I was hoping to do some actual testing of different equipment, but that really
does depend on being able to get some repeatability.