You can see my first prototype on the project photos. Good news: it works! I've run it for a day and could see how much water I use almost instantly! It was  fun. 

Problems discovered so far:

PCB coils have shortcuts

This results in less inductance and they become less sensitive. Contacted the manufacturer about it. Worse case: I have to increase the trace width and spacing. This will also reduce inductance, but keep production more reliable. 

I2C lines need pull-up resistors

I knew it, but was hoping to use internal pull-ups of CC2652P chip. No luck. Had to add them to already produced PCB manually. Improvement for the next version.

Battery dropped by 2%

This issue worried me most. I've tested it extensively with my Nordic PPK2 for many hours, but it seems to be working just fine. Average power supply current is circa 20uA - this is actually good. May be it is just cold under the floor? Have to monitor it for a longer time.

Count was 6 liters off

This one is also nasty. Took me a day to redesign the counting algorithm. If you want to know more - ask me in the comments, I can make a separate blog post about it. For now I can say that algorithm has changed significantly and I am also counting "missed triggers". There are 2 sensors and they have to be triggered one after each other. This counter is incremented when it doesn't. 

I am testing new firmware for a second day and it follows meter precisely and zero triggers have been missed. Will see what happens in a couple of weeks.