The sketch in the emonTx & emonPi is essentially the same. If the a.c. sample is present when the sketch looks, which is 500 ms after start-up for the emonPi, and after the rf test sequence (approx 1 s after the sketch starts) and as part of the initial 10 s power-on-self-test for the emonTx, then it uses the a.c. input and calculates real power. Otherwise it estimates apparent power from the current and the nominal system voltage. That doesn’t explain negative power readings (because you can’t have a negative power if you don’t have a voltage reference to determine the direction of power flow) - unless you’re doing maths and subtracting one power from another.
Usually, powering both the a.c. adapter and d.c. USB supply at the same time is OK. Powering the a.c. adapter late is not a good idea. I’ve never heard of a problem attributable to powering the a.c. adapter first, but this will stress the protection diodes in the ADC input, so it’s probably not advisable.
Going back to @overeasy’s point about commoning the a.c references - yes, you must observe the correct polarity all the way. The plug outer is GND reference on the Pi and on the emonTx V2.5 upwards. It’s only the emonTx V2 where the a.c. input floats at 2.5 V d.c.