I thought most modern devices had isolation!
From IEEE 802.3 (revision 2012), section "32.6 PMA electrical specifications":
> The PHY shall provide electrical isolation between the DTE or repeater circuits, including frame ground and all MDI leads.
> This electrical separation shall withstand at least one of the following electrical strength tests: > [...]
> b) 2250 Vdc for 60 s, [...]
Non-compliant Ethernet PHY?
Active PoE is negotiated with handshake
The problem with PPoE in these cases is, I think, not the voltage so much as the current. The continuous 24v supply may overheat the magnetic coupling transformer and cause it to fail. Some Ethernet interfaces, usually on telecom equipment and quality switches, have over current protection to prevent this. Unfortunately consumer devices usually don't.
It's important to understand this because 802.3af etc. does provide power without being asked - as a rest for a characteristic resistance on the receiver. Otherwise it wouldn't know if a PoE-capable device was connected. Up to 20v can be applied during this process but it is time limited. In general, 802.3 PoE supplies must monitor the current usage of the powered device and cut off power if it is too high or even too low for more than a short period of time. This is in part to prevent this overheating problem on devices that might, for some coincidental reason, fall into the appropriate resistance range to activate PoE.
In other words, 24v or even hundreds of volts for a few seconds is perfectly safe. 24v for minutes is likely to cause damage to devices without better protection than the spec requires. Old Ethernet equipment used to make the non-isolated components relatively easy to replace so that repairs after a problem like this were easier but now the isolation is a tiny surface mount part and replacing it will require tools and skill.
Granted, CAT 5-8 cables are supposed to be twisted pairs and sometimes shielded, but it's imperfect.
I wonder if the problem is that 1:1 signal transformers for decoupling are being replaced with simple DC blocking circuit. That can be most simply done with a spare 0.1uF and an R10k per pin, which by the way generate load of from V/R=I 48V/10kohm = 5mA and 48V * 5mA = 230mW > 1/6W. That could cause resistor to burn off if phone-sized components would be used. Or if the cap might only be rated for 10V, it could burn open. I have nothing to support these hypothesis though. I could be entirely off.
Also I suspect the "passive PoE" mentioned above could be cheap injectors with very rounded corners and always active 48V. Those are widespread for surveillance cameras and other neckbeardy applications.
There's three common pin outs. You've described 'alternate B' -- the unused pairs of 10/100; you can also use 'alternate A', using the pairs used for 10/100. Or 4-pair would have both.
My personal involvement on this was several years ago when I was going to buy several unifi access points. It turns out you have to be careful because many of the models advertise as being POE but in reality are jank 24v passive systems. I have not kept up with the current unifi lineup but at the time you had to make sure to get the "AC Pro" to have real 802.3af compatibility.
I don't know how true it is but I heard horror stories where ubiquiti 24v gear would activate, then fry when hooked to a real poe source. personally I suspect it was a 48v passive source.
I did a quick informal survey of the ui store and it looks like the newest generation(U6) all handles 48v while the prior generation(AC) the pro models were 48v and the lite and long range models were 24v passive.
By older I mean you could still buy it three / four (?) years ago
It's usually magnetically coupled. I've only seen capacitive coupling when both devices are on the same PCB