The XYZ values should not just be written to those addresses.
This is what should happen, if I remember correctly...
Read XYZ values from sensor. Multiply those values by 2 as the offset register has twice the resolution of the G readings. Read the current XYZ offset values out of the chip. Add the current offset values to the newly calculated values. Write the combined values to the offset register. Repeat.
Also note: The value for Z should be offset to 64, not 0, so use (Z-64) as the value for Z. I can't remember if the XYZ reading should be negated before being added to the old offset value.
I received my board one or two weeks ago, from SeeedStudio. It still had an old firmware that had RLE bugs.
If you disable RLE then your signal might look better.
Also make sure that you are using a new client. I suspect that the client I started with also had RLE bugs...
Between the firmware and client bugs I got the same results as you: Grounded pins still had noise on them. No understandable signal was visible... I also thought my board was broken...
