I've been pretty happy with the metrics from my GPS for noticeable climbs, but started to suspect it a couple of months ago while scouting training routes. It seemed to read suspiciously low for small features. I didn't care much at the time and just found a different place to cover elevation changes.
Today I got a bit bored doing a long, but easy ride and took the time to see what was happening on small features; i.e. little rolls 2-3 stories high. Over about 3 km of these small rolls, I got credited with climbing 3m - clearly bogus and a reproducible result. This was at a pace of about 25km/hr. I found a dip of 4m under a bridge at the river and did some testing.
At 15 km/hr, going down the dip, up the other side and returning credited me with 1m. I slowed down to 10 km/hr and watched the current altitude reading, which dropped by 4m and then came up again. The climb counter, however had its own opinion and declared a total climb of 3 m. I went through this routine several times with similar results.
In the grand scheme of things, it isn't a big deal. Most people here are going to use real hills to work on their ascents. But I thought it was interesting, particularly on a weekend where I had specific ascent objectives.
I still trust the big features measures to be close enough for my needs.
I am not an expert on this but I believe when you upload gps data if you have elevation correction selected it corrects the elevation recorded on your device using another data source that is more reliable. Not sure what this source is but I have found the climbing results that show up on AP are reasonable. If I want to be more accurate I will use an altimeter to measure elevation change but have come to except the GPS correction on a regular basis.
Others may be able to provide more insight. If not try posting on the orienteering forum. They know there maps better.
I've never done testing like you described, but I've been road riding with a couple guys this year, and we always have drastic results. One guy rides with a Garmin 800 (810?), one with a forerunner 300, and me with my Iphone Strava app. The 800 series garmin is always way more elevation - but I actually thinking too much. For example, yesterday's ride, I had about 150 meters of total climbing, and some of the other Garmin's are closer to 400 meters climb. I actually feel the Iphone is the most reliable, as it uses data if the GPS signal gets weak. Also, I used in on measure climbs last year in France, it was always bang on. My only conclusion is that altitude data is not reliable within a certain range, unless you're doing known routes with proper measurements.
I never trust GPS. I always felt and have proved to myself that a barometric device is much more accurate. Here is an article that has way to much info about it.
http://www.dcrainmaker.com/2010/05/understanding-s...
That was an interesting link. I understand the GPS problems, particularly if you are near noteworthy things like established roads and railway tracks where you would expect the altitude to be known.
I haven't done a very careful analysis of it, but I believe I've observed anomalies in barometric devices - things like the altitude being different at the same point a small number of hours apart, or traveling in the opposite direction with (then against) a strong wind. I might craft an experiment for this to do some weekend when I'm bored.
Just casual curiosity.
I am using a Garmin 910xt which has both GPS and barometric measurements, and just for good measure AP corrects it as well by some magical process. I think it's as accurate as it gets, but not sure just how accurate that really is.
I have used barometric altimeters in the past, and they need frequent calibrating for temperatre and weather changes. GPS alone is often way off.
On the subject of small climbs, I often ski laps on a 1km course with ~15m difference between the low point and the high point. Over the last three years, I have noticed very consistent quirks in the elevation profile that AP shows - the first three laps in a session are invariably smoother (with less difference between the elevation of the high and low points) than the rest. I use a Garmin Forerunner 305 with AP's elevation correction.