Detecting sleep start

It may be wishful thinking, but it seems like there is some learning after editing the sleep start and/or finish times. Initially it would mark getting up in the middle of the night as the end of my sleep period, but I would look at the movement graph and set it correctly. After doing this a few times it seems to get it correct now more often.
My related complaint though is that the system is too lenient counting lying in bed awake as light sleep,

Slightly off topic but I don't get what use sleep info is. I go to sleep and I wake up, if I go to bed earlier I get more sleep.

On topic, sleep tracking has always been hit and miss and likely always will be. Ignore it

We are all different! ;-) I am interested in the total sleep I get and the average over time. There are days it takes quite a while to fall asleep, other days almost instantly. The same applies to waking up. The different sleep stages are less interesting for me, since I am not sure they are detected correctly.

I owned a Fitbit Charge HR that was spot on detecting sleep start and end for me. The Garmin VA HR I used after that was terrible (almost 2 hours off on an average). After one month with the FR645M I get the impression that it is slightly better, probably because now HR data is also used for sleep detection.

I posted this here, because as a software developer myself, I cannot understand how this data (when you look at the movement data in the graphs), can be interpreted this way. This is not a watch issue, it is Garmin Connect and the algorythm it uses for sleep detection. Even with the newer watches with the elevate HR sensor, the algorythm is still flawed it seems.

I had a Misfit Shine 2 (accelerometer only) and successful sleep tracking with that device was a big reason for being interested in spending money on a better fitness tracker. The FR645M does not do quite as well as the £17 Misfit did, but often I can adjust and may write an app to do what I want.
The Shine 2 reported deep, light and awake, and I think the processing was all in the iPhone app. Deep segments were spread throughout the night, and it was unforgiving when I was awake but lying as still as I could - it always seemed to label those times as awake.
The significant value feature for me was realising - as I found confirmed on their forums for other users - that 5 hours of deep sleep was the key to feeling rested/happy the next day or not. Total sleep hours, whatever else, didn't matter - just a total of approximately 5 hours deep sleep in the night and I am good. I can approximate this with the Garmin results adding deep and light sleep totals together, but as noted it doesn't adequately discard the periods lying awake (although it did do last night) so this calculation doesn't always work.

I hate to be a party pooper here but the light, deep and awake are guesses. I would imagine your deep sleep and feeling great is a placebo.

I get the same with performance condition when running, if that shows a good score I suddenly feel great, it then falls 5 minutes later and all of I sudden I start feeling tired. I turned that off.

I get that we are all different but you can't affect sleep, you can't train to deep sleep more. You can train to go to bed earlier and don't need a watch for that.

I agree with you about the sleep labels being relatively meaningless, and certainly REM can not be determined from pulse and wrist motion - but now I have the movement trace on the Garmin it is pretty clear that the Misfit 'deep sleep' periods are where the movement trace is flat-lined. I suspect the heart rate info just confuses things, but will hold judgement on that until I write some code and see for myself (there's a Vivactive HR app for which the author seems to contend that HR data alone is sufficient for sleep assessment). In my case the Garmin view seems to be that the deep sleep label is something weighted toward the beginning of the sleep cycle, rather than periodically through the night as the Misfit algorithm assigns.

I also agree there's a psychosomatic aspect to seeing the result, which made it more interesting to me to find the independent post on the Misfit forum matching my experience. It's my observation that some individuals sleep more efficiently than others - whether this can be trained or not, it is nice to have external validation that I should stay in bed longer :-)

It's funny and interesting to see what people use these watches for and the interest they bring from understanding the data.

I haven't looked at sleep data for months before reading this thread. To my amazement the data actually looks pretty accurate from a fall asleep and wake up stand point.

For me though it doesn't tell me anything other than I get 7 hours sleep which I know and what I aim for.

All very interesting to read.:-) I think each person will value/validate this data differently, but... we got a bit off-topic here.;-) My issue is not HOW and IF to interpret the sleep data itself, but how Garmin detects sleep start. For some people it does seem to work, and not for others. But looking at it in a very simplistic way, if you look at movement and HR data, how can Garmin say sleep starts when there is still so much movement and a certain (higher) HR level?

But looking at it in a very simplistic way, if you look at movement and HR data, how can Garmin say sleep starts when there is still so much movement and a certain (higher) HR level?

You're correct that it shouldn't start sleep while detecting the higher HR and movement. Garmin does some things well and other things poorly. I was interested in my sleep patterns only because if was available. I realized very quickly that what Garmin was reporting was very inaccurate. I don't even bother with that data anymore. You yourself have stated the numbers don't make sense. You know what your body is doing better than your 645. Take the data with a grain of salt. It's just my opinion, but I don't see Garmin's sleep tracking improving.

In a complete 180 my data doesn't reflect yours. It puts my sleep at the time when there is very little movement.

As to your question, my guess it has to do with sudden drops in activity. There is a big rollercoaster type drop in both of you pictures where it appears you stopped moving. Unfortunately it appears that you were just incredibly still. The watch presumed sleep but you then subsequently move.

I think my pic shows this is likely how it works, the difference being that I wasn't totally still prior.