Stepper Motor Machine with HOME_USE_INDEX proximity switch?

What is the minimal setup required to use HOME_USE_INDEX assuming a stepgen and no real encoder feedback? I would like to increase the accuracy of homing while maintaining my beefy coolant-proof limit switches.  Their repeatability suffers with slime and chips ( /- 0.005 on a bad day).  To that end I have added a proximity switch on all 3 axes and written some "special" GCODE to hunt for a slug on my leadscrew.  I hunt for it bidirectionally to compensate for temperature induced sensitivity changes.  I do this with some custom M-code and GCODE. This works great but can get in some funky states and sets a few brain-fart booby traps.  It also might get me nominated for a The 2022 LinuxCNC Rube Goldberg Award (which I would like to see as a real annual thing!  Maybe we use "user says thank you"'s as votes? )I use "seek to switch and then back off slowly" for my homing currently.  To get my fancy prox switch stuff all in HAL, I'm writing a "stepperprox" component that during homing will latch the home switch until it gets to the calculated center of the prox target.  I would totally be okay with being told how dumb this is and what the simple solution is. Thank you!       

This is the third time i am reading this and still failing to see how would a slug on the screw shaft be more precise than the encoder attached to it with 500 or 2000PPR ? If using linear scales that becomes even more weird.
Also, proximity sensor is how precise ? I doubt it can keep up with the industrial one.
But i do like the "measure from both sides and calculate center" idea, nice one.

Still no award for you, in general this is a good idea, the sensors might not be up to the task.

@tommylight --  Sorry the post was confusing. I don't have encoders or servos.  Just the steppers. The slug on the screw is to compensate for that fact.  The prox is kind of like having only the index channel (Z).   I'll keep trying for that award   The dual approach gets me from 0.0015" of potential offset to about 0.0001", so more than 10x better.  I used a heat gun to raise them from 60F to almost too hot to touch to compare the single approach to the double approach.

Yeah, now i recall reading that,thank you.
So we are back to "good idea".

yeltrow wrote:

 I would like to increase the accuracy of homing while maintaining my beefy coolant-proof limit switches.

You might be in luck. In LinuxCNC "master" (the dev version) the homing routines are now a pluggable module. (thanks to Dewey) 

github.com/LinuxCNC/linuxcnc/commit/08ac...b0cd47dc0eedfbc36785

That sounds like it will be great.  I will be sure to check it out.  I have something working for 2.8 that will need re-written for 2.9 to use in that way.  The machine that I use linuxcnc on is production machine, so I'll have to plan my migration carefully.  I read through linuxcnc/tests/README .  Between that and the abs.comp tests you pointed me to, I hope I can produce something of quality.  Thank you.