Piece processed with compensation files activated in linear axises

Hi. Recently i've processed a piece, which looks like stairs to measure manufacturing errors of my machine. Before processing the piece, i've measured linear positioning errors in my linear axises(X,Y,Z) and fixed them with compensation files. After i observed my linear positioning errors has been fixed, i've processed this piece. Before compensation files, i've processed a piece which has length mistakes in linear axises like X u. After compensation files, i've sent this pieces to CMM and the results are even worse. The mistakes have been like 2X in same piece.

Here is my questions,
1. Why is it happened? What did i wrong, or is there anything that i should also do after measuring linear positioning errors?
2. Is it possible this problem is related to B_TABLE_OFFSET variable? What happens if B_TABLE_OFFSET variable has been set wrong?
3. What are the possibilites that makes these errors increased instead of reduce? Because when i looked at the results, it seems like the errors have been doubled for horizontal and vertical axises. Is there any posibility like this, for example i made reverse compensation, or i've entered wrong sign coordinates to comp files etc. ?

Thank you for your answers,
Berkay

Interesting. Impossible to say what's happening without knowing the tolerance, range, setup, how you measure the screw error let alone how you validate the interpolated comp data you are using. All good guesses... 

I use voodoo & telepathy for my comp, and put up a steady 0.01~0.02mm tolerance across the entire screw - as in same tol as what the linear scale tolerance I use to measure the error is certified/calibrated at, naturally. Fully automated using a 2D array (python) and a pulse counter (stm32). I use a zig zag measuring pattern, see code below if you need to make any sense of what I'm trying to say... 

Thank you for your answer, @Mecanix. I don't understand what did you mean by 'vodoo & telepathy'. Can you explain what is the context of these words?

According to my measurements, tolerance of my setup is over 0.08mm. I measured this by comparing different measurements with each other. So in summary, i think I've avoided the mistakes can be sourced from movement or vibration of the table which setup is located at. I am using linear scale which has a tolerance about 1um, and also comparing the results with digital comparator before starting the measurement process. There isn't so much differences in two results.

For the future, i am also planning to automate this process by running only one gcode and getting the results in csv or other formats, but without verifying that i am able to reduce positioning errors with this method, it is postponed for now.

If you explain the vodoo & telephaty words, i am gonna check out what can i do for this.

Thank you again.

In addition, i want to ask you what is zig zag measuring pattern and to upload the attachment again. The file that you uploaded is not working. I can't see the content.

Thank you.