HAL for open-loop encoder homing

For reference, here are some results showing a comparison between the original proximity switch homing and the "index gated" implementation.
 
The "optimisation" process produced the biggest improvements which were the result of finessing the PID values and slightly reducing the homing speed to minimise the delay between reading the index pulse and processing the data. Both methods were significantly improved this way but the end results were almost identical.
For the test setup, I repeatedly homed the machine (typically 20 times, as shown), with the absolute position automatically logged and plotted.

It sounds like you are not using the index hardware at all.
There should be no effect of LinuxCNC timing or PID tuning
on the index position homing precision, if the index hardware is
actually used.

That is, the precision of position capture at the index position is
independent of the motion accuracy, though of course motion accuracy
after the index position is captured _is_ dependent on PID etc.

I'm certainly open to being educated here but I was reasonably convinced by what I was seeing. Have a look at my notes and perhaps you can see where / if I may have misunderstood what was happening:
www.murraye.com/2026/08/improving-homing...acy-on-linuxcnc.html

In my case, with 5um resolution scales and closed loop controls, my final resting position after homing can't be any more accurate than the resolution of the scales allow, so perhaps I've been chasing up a dead end anyway? A positional accuracy of 5 or 10um from true is probably the best I could hope for on a good day.

The PID element was initially due to my controllers being unstable but the homing velocity did seem to impact accuracy too.

I think the issue is that you are not using the feedback from the same encoder that's generating
the index. In this case, homing to index cannot work.