10K was short, OK to conservatively convert?

How is a GPS watch not accurate? I run with mine all the time and my distance is always the same if I run a familar route where I have previous runs to serve as data comparison. Only time my watch ever freaked out on me was when I ran underneath a bridge.

One GPS measurement was 5 meters short and the other was 8 meters long. The calibrated bike measurement was 7 meters long. The fact that one GPS measurement was long and one short, with a difference of 13 meters, is actually a good thing. Unlike systematic error, random error can be minimized by taking multiple measurements. The average of your 2 GPS measurements was only 1.5 meters in error.

I'm not suggesting that GPS measurements are more accurate than a calibrated bike, even on a grass course. You got lucky with your 2 GPS measurements. I don't think there's any commercial GPS device that consistently gets that close. Errors in the range of 0.3-0.5% are more typical for the better ones. I'm only saying that your experiment does not suggest the GPS you used is inaccurate, or even that it is less accurate than the calibrated bike on that course.

But even using your standard of repeatability as the measure of accuracy, 0.27% repeatability error is not what most people are talking about when they say GPS is inaccurate. Multiple blogs and web articles talk about GPS "always" or "typically" giving measurements that are >1% or 1-3% long. People who take GPS measurements during races often get measurements that much too long because they don't follow the SPR, or they went off the course to the port-a-john, or they didn't give the device enough time to acquire satellites before the start, or whatever. But the devices themselves are not that inaccurate, and your experiment is more evidence for that.