How is the watch for gps accuracy on MTB trails in deep forest

The watch is very accurate for MTB from most accounts, including my own. But how would you really know? I mean, what is the distance against which it is being compared? On the road, measuring true distances is difficult but possible. On MTB trails, it's pretty much impossible. Lots of people use Fenix watches for MTB on Strava.

Fair, I guess one method would be to assume trailforks data is accurate and compare against that. Or does starva measure segment times from two specific gps coordinates? The main goal is to get accurate segment times rather than accurate lines I guess, if that makes sense

That may be a good assumption - I have no idea. But for me, it makes as much sense as assuming the Fenix distance is correct and evaluating trailforks based on that. Assuming one measurement tool is accurate does not make it accurate.

If you’re riding the same segment multiple times you’ll get a feel for the distance. You could always get a measuring wheel and check of course. 
FWIW, I ride with several people on trails in heavily wooded sections on gnarly singe track occasionally. We’re never too far apart in the distance recorded. And that has changed little with either the F7/Epix. 

I recently got a F7X SS and did a reference hike with it, as measured on the ordnance  survey map 1:25000 by creating a route on a PC, it was 9.71 km. As measured on the F7 it was 9.74 a difference of 30 meters. All satellites, smart recording, right wrist, 3K at least under tree cover.
I was pretty impressed with that.

We don't have deep forests around here, but somewhat related:
Returning to squash after injury, I went and hit a couple of balls the other evening, yesterday I played a very light game. I used same as in F5, an adapted "Walk" function to record. May try tennis next time. My F5 used to give me tracks outside, shooting off into the next door river, etc. 

F7, dead solid within the courts! And much more realistic lower distance. Despite how you feel, you don't run 5k on a squash court during a game! So even indoors the GPS was solid in my experience. 

If you want to check the distance recording for cycling, buy a bike computer for 20 bucks, callibrate it correctly on a road and then compare the data after a trail ride.

That's a good idea. It does require that you take exactly the same line on your reference ride as you do on your measured ride. With road riding, that's not difficult, but with a MTB trail of any significant length it's quite a bit tougher.

It comes down to how accurate you need to be.

What do you mean with reference ride?

A classical bike computer measures the turns of a wheel and calculates pace and distance from that. Once calibrated correctly (just measuring the wheel diameter is not sufficient), it is the most accurate measurement of your distance you can get. You can use it on every ride and compare the data to the ones measured with GPS.
BTW those computers were already available in the 1980s, so correct measurement of the distance was common long before GPS was available.

You can also use a bike sensor and connect it to your watch, but AFAIK it is calibrated via GPS, so it might not be as perfect as a perfectly manually calibrated computer.

Might be most accurate for road where a lot of the variables remain the same - and the wheels are pumped so hard that it should not be noticeable.

The following needs to EXACTLY the same, most of which is hardly ever the case on a MTB ride:
Tyre pressure, suspension settings, rider weight, kit and bottle weight - all of this affects the diameter of the wheel. 

Then, as mentioned, exact same line through every turn. No wheel-spin or lifts or jumps, no slides, no locking of brakes.

To the OP: Strava usage and segment times is mostly based on start and finish, given that you don't shortcut it. It was built to initially work with their own cell phone app. I've not had any issues with Strava and Garmin integration - except that I'm slow.