NOTE: This thing is definitely NOT WORKING, the board is moving and so is the sensor... Gluing the sensor does not help, Heck the whole effector is moving sideways left and right this can never work reliably, i don't know why i though it would. You always end up with the hotend hitting the glass before the sensor could pick up the magnet. This is because for some reason the effector is not staying leveled. Slowing it down does not help. As long as you have a force moving it down, the effector is gonna wiggle a bit on one side or the other.  I really hate having to go to the FSR route because with my bed that's not possible(too much tape...), i will need to buy one of those Chinese pcb heater integrated with an aluminium plate but that means another 2 months of waiting for the package... i use this with 4 10x10x1mm magnets, i had these lying around, hidden in a software box.i use M4 nuts around my bowden tube because it's stronger than Push-in, and that's also what is on my Hercustruder.i've been able to use a spring with this using a big metal bolt and a piece of foam tape on top of the magnets to prevent the magnets from sticking directly to the bolt. i'm using an old marlin because the latest ErikZalm's version still does not support delta auto probing. i would have much better result with slower and repeated probing. And also more consistent result if i had hall sensors instead of switches for homing on the 3 towers.http://youtu.be/u-wHDbS9njQhttp://youtu.be/g1FcLM-XEyEAfter these video, i did find i could slow down the probing by modifying a line in marlin_main.cpp, so if you want to do the same, look for:static void run_z_probe() (a few line after change this line: feedrate = homing_feedrate[Z_AXIS]/4;for this:feedrate = homing_feedrate[Z_AXIS]/8;Note that i'm using jcrocholl's marlin and my copy date from before the FSR mods.