# Sense Threshold for Always On?

Shouldn’t Sense only place devices “Always On” that are on 24 hours a day?

What is the threshold?

Or is this a bug?

How do I remove this device from “Always On”?

You must have your Garage Fridge on a smart plug. Three things you need to understand:

• Sense analyzes the power usage of every smart plug to determine the rolling Always On component for each.
• Sense also uses a user settable standby criteria (power time threshold), so it can actually track on, off and standby. You want to set a standby criteria that is higher than the Always On for that device, or Sense will never see a standby level for that device.
• The Always On for all the smart plug devices is shown in the Always On “Device”. You can even use the results from the smart plugs to reduce the unknown part of the Always On.

1 Like

But the device is not always on a can be seen in the 2nd pic.

Pretty sure it’s set correctly.

Ah, good question - Always On is a statistical calculation. So the smart plug can be at zero for some small part of the time and still register an Always On value. 1% of 24 hours is 14.4 minutes. Given the way Sense phases the definition below (based on estimated distribution) there might be a little bleed-over beyond 14.4 minutes, so I’m guessing that a device that is at 0 for less that 20 min a day and well above that for the rest, is going to have an Always On.

Always On component of a smart plug device
This is an estimate of the Always On-ness of a smart plug or dedicated circuit device based on its power draw over the past 24 hours. It is updated once every few hours based on consumption data. Always On-ness is determined by looking at the distribution of wattage at a high granularity (1 second) and estimating a value corresponding to the 1% bin of its observed wattage histogram.

ps: If you want to get into the discussion of estimated statistical distributions, feel free to ask. The main house Always On uses a 48 hour discrete selection of the 1% bin rather than using an estimated distribution. But a while back I did a simulation of a 48 hour estimated distribution Always On value based on 3 different normal distributions of data 500W average with a standard deviation of 10W (blue), 100W (yellow) and 200W (red). I then fed in an increasing number of 0W, to see when the estimated distribution 1% bin hit zero.

1 Like

I’d like to explain this another way. Let’s imagine you have 10 lights, 100 W each, and each runs 24 hours per day except for a single 15 minute period. Assume each light goes off for the one 15 minute period on some random quarter hour. In your view, @Beachcomber, the “always on” for each individual device is 0 W, so the entire house has 0 W always on. But the chance of all lights being off for the same 15 minute period is (1/96)^10, i.e., near zero. In fact, the chance that two lights are off for the same period is under 62%. From the total load view, the “always on” load is 900 W. If only one light (maximum) is off at a time, you have 9 x 100 W on at a minimum.

3 Likes