03v6 is the hardware version (yes, confusing, should have separated the hardware and firmware repos). Do you mean that you used the release called "Hardware update, 03v6" (tag v2.5) or "New firmware implementation (v3.0) (tag v3.0)" ?
see it reporting voltages of 0.00 in the chunks, which indicates that it's using firmware version 3.0 and not 2.X. But when they send data to the server, it says the last voltage is 2.880. I don't remember the protocol, so perhaps that's ok. But it does seem to indicate that you are using the newer 3.0 version. You can tell which version it is based on how it blinks in when you turn it on - version 3 does (I think) - Red on for 100ms, and then blinks the green led three times.
My current guess:
* An unknown bug in V3 of the firmware that we don't know. I did all of my live experiments with v 2.6
* Data corruption on the badge. Either because there was old data there from v2.6 or, something went wrong. Wiping the memory is a bit tricky ... I can't remember how.
I'd switch the "broken" badges with backup badges. You can do this by changing the MAC addresses but keeping the same name. This way, the new badge will get the same "observed ID" so the analysis is easier. Don't forget to turn off the broken badges because they'll keep broadcasting the ID otherwise.