Why does my motion sensor keep flipping off when my kid sits still on the couch?

What you're seeing is basially baked into how this tech works, so I wouldn't chase it with placement changes. The sensor reads disturbances in the Wi-Fi signal between your board and the router, and a walking person makes a big messy disturbance. A person sitting still makes a disturbance for a few seconds, then the signal settles around them and they just become part of the baseline. At that point the score drops and it reads as an empty room. It's motion sensing, not true presence sensing like a mmWave unit that can see a stationary chest breathing. That's why bumping the off threshold and the off hits only stretched things out instead of fixing it. I ran into the same thing with a chair in my living room, you get maybe a minute or two of grace and then it clears no matter what you set. And the router sitting between the shelf and the couch acutally makes it worse in a way, beacuse the constant traffic on that path drowns out the small changes a seated body makes. Worth trying a recalibrate with the room in its normal state, so the baseline includes the kid on the couch rather than an empty room, but don't expect miracles there either. Moving it closer or re-aiming the antenna will change sensitivity a bit, mostly for range and thru-wall behavior, but it won't make a motionless person read as present. My take after living with it for a while is that it's genuinely good at one job, catching someone moving through a space cheaply with hardware you already own. For the couch scenario I ended up pairing it with a cheap mmWave sensor on the same board, motion from the Wi-Fi side for the wide trigger, mmWave for holding presence in the seating area. If you want to stay all software, you can stack two boards so one covers the couch area from a different angle, wich helps a little when someone shifts position, but it still won't hold a truly frozen body. So yes, it's how it behaves, not your board.

Mine does somethig similar with the ceiling fan. If it's running on low the score settles right down and reads clear, then the moment someone walks under it things go nuts. Took me a week to figure out the fan blades were eating the baseline. So yeah, anything that moves air or wiggles the signal path quietly can throw it off in ways you don't expect. One thing that helped me was keeping a small motion trigger on the entry door, so anyone coming in bumps the score even if they go straight to sitting. Doesn't solve the frozen-on-the-couch problem but at least the lights don't die on arrival.

One trick that worked for me: stop relying on the sensor alone for rooms where people sit still. My TV room lights are held on by the media player state, so as long as something is playing the lights stay up and the motion side just handles entry and empty-room shutoff. Took about ten minutes in the automation editor and saved me from buying anything else. The score graph is still handy for tuning the entry trigger though, I found walking speed matters alot, a slow shuffle barely moves the needle compared to a normal pace.

Update: the media player trick worked, lights now stay up while something's playing, so movie nights are fixed. But the sensor itself still reads clear when he sits still, which makes the empty-room automation fire early in the afternoon. Does anyone here pair it with a proper presence sensor to cover the sitting-still gap, or is that overkill?