shutting off laptop
... I take it you mean 'shutting PC with suspect PSU ?
- you can see that a 1.5db drop brings it down considerably & close to what I think may be the 'limit' before - is it the dslam??- decides enough is enough & reduces speed to improve SNR.
The SNR should be able to drop to 9, and well below it, before anything gets disconnected. Some people even report brief spells where the SNR has gone -ve, and survived.
However, if your SNR was (say) 8dB, and you got a short burst of interference that reduced it by 9dB then you'd probably drop the connection. I'm not sure how long a burst of interference would be required for the connectiuon to drop, maybe some of the Gurus can shed light... a second? more? less?
Be aware that such a burst of interference wouldn't necessarily show up on routerstats, which just takes some atomic samples at regular intervals, and joins up the dots. If a burst of interference occured between samples, routerstats won't show it.
This is making me think, however... didn't you previously report routerstats sometimes showing sharp dips, which we probably attributed to a glitch in the wireless connection, or something? Suppose these dips were real, and happened whenever routerstats got lucky, and took its sample just when the intererence burst was happening?
If you still have the energy to investigate, you could try reducing the routerstats sample time, which would increase the chances of a sample being taken during 'bursty' interference? I've many times confessed to not being a routerstats expert, so somebody who is may correct me.