To answer #3: If you look at the ublox datasheet for the F9P (see the Facet product page at Sparkfun, the datasheet is there); ublox specifies an RTK accuracy of 0.01m + 1ppm. The “1 ppm” is the error added by the baseline length. (That’s the distance from your rover to the base GNSS receiver. Note that the caster is a computer server on the internet that (broad)casts the data to your rover, it’s the distance to the base GNSS receiver that is important, not the distance to the computer server.)
Surveyors have a joke about accuracy:
Q: What’s the difference between a used car salesman and a surveying equipment salesman?
A: The used car salesman knows he is lying.
The moral of the joke is to take the accuracy specs with a grain of salt. Trust but verify.
GNSS equipment accuracy specs reported by the manufacturers are usually one sigma. That is, if everything is perfect, the coordinates will be within the accuracy specs about 68% of the time. And 32% of the time they won’t. And sometimes they really really won’t. That’s a fact of life, it’s not a deficiency in the F9P or the Facet.
Surveyors may take multiple shots on important points to confirm their readings. Best practices include separating the shots by hours or days to get different conditions.
You need to use proper surveying techniques to reliably get the manufacturer’s accuracy. Getting the Facet accurately centered over a mark in the ground and properly measuring the height above the mark takes practice and care.
By far the biggest issues folks run into involve their understanding (or lack thereof) of geodesy. There is a lot to learn about horizontal and vertical datums. There are many different ways to accurately express the coordinates of a survey mark in the ground. Geodesy helps you understand the different ways so you can compare the numbers taken by different people or published by different agencies.
You need a good clear view of the sky.
Bad RTK fixes happen, especially in challenging conditions. Buildings, heavy tree canopy, mountains, etc all interfere. A bad fix is when the unit reports an RTK fix with great accuracy but the unit converged on a bad solution. It’s a fact of life, not an equipment deficiency. Surveyors spend $30,000 US on their GNSS gear and better bad fix rejection is a big part of what they get for all that money. Take multiple readings and compare.
To use an analogy, Sparkfun sold you a very functional and precise car at an amazing price point. It’s on you to learn to drive it. It’s not hard but it take education and practice. To stretch my analogy, Ford and Toyota don’t teach folks how to drive, there is a lot to learn. And some folks learn to drive really well.
Good luck and have fun!