Then again, if it really is truly random, how are you going to set up a repeatable test on an ICE
you can't
what you can do is 'gather intelligence' and, eventually, based on that 'intelligence'
A. Definitely catch the whole problem - from cause to effect - when it happes, and
B. Prove that any proposed "solution" does actually fix the problem!

The only 'bug fixing' that should be acceptable is 'bug removal'. I recall a case where some i.... after inserting 'traps' in a program saw the intermitent go away and shipped the product with the traps.

No ICE, simulator or acme wonder can remove a bug, all they can do is point you towards the place where you can reason the cause of the bug. I just find the ICE often is the best 'ponting device'.

I do not, by any means deny that e.g. an algorithm can be debugged by just about any debugging means known to man, just those pesky "once a day" bugs are, in my experience, not debuggable in a 'productive' fashion by any other means.
The very best tool for debugging is, for sure, a "DNA computer", but every aid you can give that process enhances productivity.

Erik