Since there seems to be some questions about the accuracy and consistency of the initialization (homing) process. I decided to measure the accuracy on my Shapeoko Pro XXL.
I placed the .25" probe in my router and established a 0,0,0 position near the SW corner of my Hybrid Table with the probe .035" above the table. I offset a fence exactly 1.000" from the x=0, y=0 position as a reference. I then jogged z up to the stops and x,y back to the NE corner of the table. On the first set of measurements, I kept power on and did not reinitialize the machine. I jogged the probe back to the 0,0,0 position and then measured the position of the probe relative the fence and table top. I repeated this process 6 times and recorded the results.
On the second set of measurements I followed the same process, but between measurements I cycled power on/off and reinitialized the machine after powerup. Again I recorded the results of the 6 passes.
The results are that the average deviation was X=0.001", Y=0.003", Z=0.001" for the 6 readings taken without reinitialization. The largest deviation was Y=0.005".
For the case where the machine was power cycled and reinitialized after each measurement, average deviation was X=0.002", Y=0.002", Z=0.000". The largest deviation was Y=0.004".
My conclusion is that my machine comes back to the 0,0,0 position pretty accurately independent of whether it goes through a power cycle and reinitialization or not. Michael Hotchin mentioned deviations on the order of .5 mm in his post. That would be a deviation of about .020", which is far more than I saw in my tests. Perhaps my accuracy will get worse as my machine ages. But, I feel comfortable using a fence to avoid having to go through the full BitZero process for most of the work I do with my machine.
I was also curious how accurate the BitZero process is to establish a 0,0,0 reference. I decided to test this by performing the BitZero process and measuring the position of the probe relative to the corner of the work piece. To do this, I established a fence that was offset from my work piece by exactly 0.750" in the X and Y directions. I then performed the BitZero process, removed the work piece, moved the router probe to the 0,0,0 position, and measured the distance from the probe to the fence and table top. I followed this process 6 times and recorded the results.
The average deviation was X=0.002", Y=0.003", Z=0.001". The largest deviation was X=0.005" and Y=0.005". This was disappointing to me. I expected this tool/process would be very accurate and consistent. I wonder if the problem is that the BitZero tool seems to move as it is being probed, even if you try to hold it against the work piece? Perhaps others have more information about this.