Communication between transmitter and app/receiver is digital (Bluetooth) on all current units, so a transmitter error will result in “no signal” rather than bad numbers. Bad numbers are a sensor issue.
As mentioned, Dexcom accepts a “bad sensor” report if you do a double fingerstick and it’s 20% off. (I don’t know which reading they calculate the 20% from – it matters but not a lot.) Now that you can report the failure online, it’s much easier since you can just fill out the form with the numbers. They don’t accept “flaky” as a reason but if it’s flaky I can usually wait a few minutes and get a 20% off reading to report.
Notably though, sometimes it will recover from flakiness and give good readings for a long time afterward. So I wait it out for a bit, generally the one hour which is also the time to report for sensor error or signal loss.
I’ve often seen the “falling off the cliff” phenomenon. Numbers will be normal, then they drop suddenly, like from 140 to 60 in ten minutes. This is obvious from the graph but it can be a shock to get an urgent low alert and be pulling out the glucose before I realize it’s the sensor. Generally it goes into Sensor Error mode at that point, about ten minutes too late. Somehow (I don’t know how) the system sometimes recovers from this situation and goes on giving good readings for another day or two. (This usually happens on day 9 or 10. Helps to build up the sensor supply. 
)
I find one thing that pushes numbers up quite a bit, temporarily, is a hot shower. I don’t know why this isn’t widely mentioned. For me it’s often 30-40 points. While it resolves within about 15 minutes, it can temporarily obscure a fast drop.
It’s also worth remembering that the makers of strips claim they are good if a control solution test is off the nominal value by less than 20%. WHAAAA? We all know that a strip giving a 20% error is dangerous. We trust the fingerstick values because we know the strips are much better in practice and the makers are just doing CYA, but … well, but.
