It depends on the sensor, really, but usually I barely calibrate at all days 1 through 18-23ish (except for startup calibrations, of course), and have to calibrate a lot more as the sensor nears it’s end of life. I trust it enough, that I almost never test anymore during those early days. And as long as I keep calibrating during the last session, I can keep limping it along until it starts getting too erratic and loses data.
Until the very end, the data is still reliable, it just starts to drift a little. So a calibration will knock it back down to where it should be. If a sensor is being particular difficult, but still yielding data (no missing dots on the graph), then I’ll sometimes do a pre-emptive restart. A lot of timrs a restart will stabilize it, and you’ll get more life than you thought you could.
I’m on a new sensor right now, but near the end of the month I’m going to try the “no code” method, now that I’ve learned it doesn’t force you to calibrate at a given time. I think since I’m calibrating often anyway (1-2 times a day) near the end, it might just be easier to go the “calibrations required, but we’ll trust anything you say” route.
It sounds like you’re probably nearing the end. You’ve done really well with your first sensor, though. A lot of people can’t get any restart time. I think I only got like 14 days on my first sensor, but sugars were crazy then. I had no clue just how crazy! I wouldn’t give up on it yet, though. I’d suggest you stop the session and restart it one more time without entering a calibration code. I’d love to see how long you can get when you’re not fighting the internal calibration programming.
Sometimes I just get a really 
y sensor, and I have to calibrate it constantly. Those ones, I usually get Tandem to replace after a week or so, though. They drive me crazy… Tandem expects your Dexcom data to be reliable. They have zero qualms with sending you a new sensor if you tell them you had to calibrate three times in a single day because it’s too far off.