Strange behavior after midnight

[final needle]

The past few days I've noticed that from 00:00-01:00 on the energy dashboard I've had a negative "untracked consumption" of around 500-1000 Wh. I don't have any power generation in my apartment, so this is all due to some math or sensor errors.

I have a standard american split phase panel, and on the main feeds I have a shelly Pro3-EM. Those two sensors are my energy "sources" rather than using the laggy grid data. While looking into this I discovered some strange values for other sensors in that same hour. In particular, I have a constantly running vent fan that draws about 15W, and my cable modem and a small switch on another sensor which draws around 13.4W. If I look at those sensors directly, I see they list their consumption for today (20 hours into the day) at 290Wh and 267Wh, respectively. Basically, exactly what you would expect for 20 hours of use at their power draw.

However, if I hover over that first hour of the day on the individual device detail usage chart, those two devices show .13kWh and .11kWh, which is obviously more than they should be showing. Every subsequent hour they are showing either 0.01 or 0.02, which is expected.

When I look at the details of the consumption sensors for those devices I see the expected saw wave pattern dropping to 0 every midnight.

So how is the energy in the individual device detail usage calculated? Something odd is going on here

[twilit blade]

Energy panel is just using the long term statistics for that entity.

Unexpectedly high consumption sometimes comes from a sensor briefly reporting 0 for a moment unexpectedly.

Share some more raw data please

History charts, statistics charts, etc.

[final needle]

Here is the energy dashboard oddness:

here is the network plug, labelled on the dashboard as laundry room networking. The change in the waveforms was due to me adding a throttle_average filter to the sensors.

and the highlighted elements are all obviously wrong, as well as the laundry machine, which wasn't running between 12-1am,

[twilit blade]

Can you get a history CSV of network plug total energy from Mar 30 11pm to Mar 31 2am ?

[twilit blade]

sensor.network_plug_total_energy,318.2,2025-03-31T06:59:43.134Z
sensor.network_plug_total_energy,88.8,2025-03-31T07:00:43.132Z
sensor.network_plug_total_energy,0.3,2025-03-31T07:01:43.134Z

This is weird.

[final needle]

yeah, it must be the averageing filter screwing that up

[twilit blade]

Ah yeah

A total sensor can only reset to 0, it can't reset to some number higher than 0 or else that is considered as additional consumption.

So that's your extra ~100Wh

[final needle]

yeah, seems likely. I think I can remove that filter entierly, and let the sensors feeding it filter. This all started because I was trying to reduce the rate at which my database was growing due to the state changes of these sensors

[unborn sorrel]

Can someone help with this as it pulls the last days data in to the first hour of the day. I have checked the time and it all seems correct, any support would be greatly appreciated.