Sleep tracking

Yes, that's the way it works since always and not only for Instinct Solar. The results are being permanently adjusted. Garmin claims artificial intelligence refining the analysis. Personally I suspect a random generator just making up the data each time you access it 

Well, the support engineer seemed genuinely puzzled by it and he is escalating it to engineering. The real question is how reliable the data is, if it is unreliable, then it's essentially worthless to keep the watch. 

8953053 said:

Well, the support engineer seemed genuinely puzzled by it and he is escalating it to engineering.

I would not call the support guy necessarily an engineer. It was a first level support person simply not familiar with this specific topic. There are countless of threads here on the forum discussing it (at different Garmin models), and I know it was reported to Garmin many times.

8953053 said:

The real question is how reliable the data is, if it is unreliable, then it's essentially worthless to keep the watch. 

If the sleep analysis is the only reason why you purchased the watch, then yes. However, the competition is not any better. And the Firstbeat analysis built directly in the watch (like at Fenix 6), is even worse, according to some user reports. So if you want a reliable sleep analysis, buy a polysomnograph. Count with costs several orders of magnitude higher, though 

Well, sleep tracking is one of the main reason why I got the watch but fair point on competition not doing any better, so essentially we are paying for a bunch of useless sensors lol 

The basic functionality, like the detecting of the sleep/awake periods, the motion, the stress, and the HR, work pretty reliably. At least in my case. Never had any issues with that.

However, as for the individual phases of the sleep, the REM, etc. - I have strong doubts about its reliability, and the main reason is indeed the fact that the graph changes each time I access it. They could have at least made it so, that the first generated version would be stored on the server for future use. In such way we could not notice it, and could not complain, unless we cross-checked the results with the output of a professional polysomnograph. 

Frankly told, I do not think they really intentionlally falsify the data with a random generator, but it is certain that the algorithm they use is useless, if it cannot replicate the same results on the same source data, when called repeatedly.