So I set up my own test: Not quite the same thing but I had a Raspberry Pi with a temperature sensor, and I only allowed it online every 10-15 minutes or so, over the past hour. Looking at my data, I see two things that I think are bugs, that may be more pronounced in your situation:
1 - Occasional data reading (not many, but some) were missed by the platform while I know the machine was online and polling for data. Considering you are polling only once every 15 minutes, I think even a single missed instance can very much mix up your results here.
2 - The hours page seems to be 'guessing' at a value on the hour (2:00, 3:00, ...) even if the device was not on to deliver a reading then. This seems incorrect to me and I'm going to suggest that we change it. If we only have readings at 2:45 and 3:15, we should not assume what the data value may have been at 3:00 in a graph.
3 - Not reproduced with my sensors, but the repeat values you are seeing in the graph which are not consistent with what the widget value is showing.
Things that I'm not convinced are a bug, yet:
- The lack of data in the 'minutes' view, which is designed to show only the last 10 minutes of readings. If you are polling every 15 minutes, I expect this view to be empty some of the time.
I will have to ask our engineering team to find out more about communication between the Arduino and the platform outside of just updating these values. I know you are interested in saving power, but I think it would be a beneficial test to allow the device to stay online for a slightly larger window (maybe 1 minute) to see if it improves any of the behavior here. It may not and we may just have some bugfixing to do, but it would be an interesting test.